A compressive sensing-driven data collection scheme for UAV-assisted wireless sensor networks Authors: Zhe Han, Guoqiang Zheng, Chuanfeng Li, Guoqiang Wang, Yu Mu Published: 2025 ABSTRACT Unmanned Aerial Vehicles (UAVs) have demonstrated significant potential as mobile nodes for data collection in sensor networks. However, the environmental conditions in remote areas often fail to offer stable and reliable energy supply for energy-constrained UAVs and sensing nodes (SNs). Consequently, efficient energy utilization strategies is crucial for maintaining the long-term and stable operation of such networks. A novel sparse compressive sensing (CS) sampling framework for UAV-assisted data collection is proposed in this work. To minimize the overall system energy consumption in each round, we design an energy optimization scheme that decomposes the energy minimization problem into two sub-problems. The first sub-problem involves the node selection under the CS reconstruction accuracy constraint. We develop a CS-based Node Selection (CSNS) algorithm that optimizes UAV flight distance while balancing network energy consumption. The second sub-problem focuses on jointly optimizing the SNs scheduling, UAV flight duration and UAV trajectory. We design a two-stage joint optimization (TSJO) algorithm that convert the non-convex issue into a convex formulation by employing the successive convex approximation (SCA), variable substitution and variable relaxation. Extensive simulation results not only validate the effectiveness of the CSNS and TSJO algorithms, but also demonstrate that our proposed scheme has a considerable energy consumption advantage compared to the benchmark schemes. Introduction Wireless sensor networks (WSNs), a key component of the Internet of Things (IoTs), have been extensively applied in various domains, including smart agriculture, ocean exploration, wildlife tracking, and disaster prediction1. However, in remote regions such as wilderness areas, deserts, forests, oceanic zones, and military battlefields, the deployment of traditional terrestrial mobile networks faces significant challenges due to the substantial costs associated with the deployment and maintenance, which hinders the efficient acquisition of sensing data2. Bogena et al. used satellites to gather data from terrestrial sensing nodes (SNs)3. However, the high-density access requirements of the SNs and the high energy consumption caused by long-distance communication render this approach unsuitable for the majority of WSN application scenarios. To address the above issues, the space-air-ground integrated network (SAGIN) has attracted considerable research interest, which employs unmanned aerial vehicles (UAVs) as mobile aggregation nodes to enable data collection and forward gathered data to remote servers via low-orbit satellites4. Due to flexible deployment capability and high mobility, UAVs not only increase network coverage and improve network capacity, but also build efficient line-of-sight (LoS) links with SNs to improve link quality and communication rate of wireless communications5,6. Therefore, the UAV-aided data collection technique significantly broadens the application scope of WSNs and offers a novel solution for effective data gathering and network coverage in remote areas. In the related research on UAV as a mobile aggregation node to assist WSN data collection, researchers proposed different solutions for various application situations and optimization objectives, respectively7–9. For WSNs with long-term monitoring tasks in remote regions, due to the limitations of environmental conditions such as the wild, mountainous, oceanic and military battlefields, the energy-constrained SNs and UAVs cannot obtain stable and reliable energy replenishment. Efficient energy usage strategies are of great significance in extending the working time of the network. Therefore, under the premise of effectively executing data gathering tasks, research on reducing the energy consumption of each round of system operation has become a hot topic of widespread concern. In recent years, the energy consumption of UAV-assisted data gathering has been the subject of corresponding investigations by numerous scholars. Regarding research on the UAV energy consumption, Ghorbel et al.10 designed a UAV hovering position optimization approach to reduce UAV energy consumption in a clustered network. Zeng et al.11 utilized the sequential convex approximation (SCA) approach to reduce UAV energy consumption during UAV-assisted data gathering with the “fly-hover-communication” scheme and the “fly-while-communication” scheme, respectively. Solar energy, as a clean energy source, can provide a continuous energy source for UAVs. Zhang et al.12 optimized UAV flight trajectories to address the multi-objective challenges of data transmission rate, UAV energy consumption, and coverage fairness by utilizing hybrid solar power and charging stations. Regarding research on SNs energy consumption, Zhan et al.13 studied the issue of node energy consumption under the time-varying channel fading model and proposed a joint optimization scheme of node sleep scheduling and UAV trajectory to minimize the maximum node energy consumption. Under the constraints of data gathering at all nodes and UAV flight distance, Baek et al.14 optimized UAV hovering points and flight trajectory to maximize the minimum residual energy of the nodes by utilizing the Voronoi diagram. For the UAV-assisted base station to provide the mobile data gathering service, Mondal et al.15 utilized reinforcement learning mechanism to reduce node energy consumption for the sparse coverage scenario of ground base stations. To minimize the weighted energy consumption across overall SNs, Yang et al.16 leveraged the Lyapunov optimization theory to optimize network resource allocation and UAV trajectory simultaneously. All the aforementioned investigations concentrate on reducing energy consumption from a single perspective. However, energy depletion of either of the UAV mobile nodes and terrestrial SNs will lead to network data collection failure. Therefore, for data gathering in remote areas with limited energy supply, reducing the comprehensive system energy consumption in each round becomes the key to extending the network working time. The UAV-SNs energy consumption is decreased in17 by simultaneously optimizing the 3D trajectory of the fixed-wing UAV and the node transmission scheduling strategy. However, their investigation exclusively considered network scenarios with sparsely distributed nodes. Zhu et al.18 proposed a deep reinforcement learning scheme in the clustered network topology with the goal of reducing the UAV-SNs system energy consumption by optimizing both UAV trajectory and cluster head selection. But SNs in the terrestrial network had to consume additional energy to establish and maintain the clustered topology. Lai et al.19 designed a UAV-assisted hierarchical network data collection architecture and developed a particle swarm optimization approach utilizing knowledge transfer to reduce the weighted sum of SNs and UAVs energy consumption. However, each round of data collection requires the participation of all SNs and does not consider the data correlation between SNs. It is noteworthy that the sensory data collected by SNs within WSNs usually exhibit significant spatiotemporal correlation characteristics. By exploiting these correlation features, the energy consumption for data gathering can be effectively decreased20. The compressive sensing (CS) technique, which enables simultaneous data compression and sampling while efficiently compressing correlated data at a rate below the Nyquist sampling threshold, has been widely adopted in WSNs due to its significant reduction in data gathering energy consumption21. Ebrahimi et al.22 utilized the CS technique to compress data on intra-cluster nodes to decrease energy consumption during data transmission, and UAV sequentially visited cluster head nodes for data gathering. This work aimed to decrease the network energy consumption of intra-cluster SNs, but maintaining the network topology still consumed SNs energy. The sparse CS technique provides a new insight to further reduce energy consumption, which realizes data reconstruction of the whole network by collecting sensory data from some SNs23. However, its reconstruction error is closely related to the design of the measurement matrix24. Some studies have proposed effective design schemes for the sparse CS observation matrix25, but these schemes do not apply to our case because the design of observation matrix is related to the UAV energy consumption in our work. Moreover, since system energy consumption is closely related to UAV flight control and node scheduling strategies, existing relevant studies have not fully exploited the flexible flight advantages of UAVs to investigate the joint optimization of UAV flight time, UAV flight trajectory, and node scheduling. To achieve system-level energy consumption optimization and effectively balance the energy consumption of the UAV against that of the SNs during data collection, we formulate an energy minimization problem for a UAV–SNs system tailored to UAV-assisted data gathering in wireless sensor networks (WSNs). Furthermore, it proposes an energy-efficient UAV-assisted data collection scheme based on sparse compressive sensing (CS) techniques. The core innovation of this scheme lies in integrating sparse CS theory into the UAV-assisted data collection framework: by visiting only a subset of CS-sampled nodes and employing high-accuracy CS reconstruction algorithms, the scheme enables full-network sensing data recovery, thereby abandoning conventional sampling paradigms that either require traversing all nodes or rely on predefined clustering structures. Building upon this framework, the optimization problem in this paper is formulated as a joint optimization of observed node selection and UAV flight strategy, aiming to minimize the total energy consumption of the UAV–SNs system under the constraint of achieving the required CS reconstruction accuracy. The following is a summary of our primary contributions. By deeply integrating signal processing theory with intelligent optimization methods, this work formulates an energy minimization problem model for the UAV–SNs system based on sparse compressive sensing (CS)-driven data collection. To overcome the difficulties arising from the non-convex problem and the absence of closed-form expressions, an energy optimization scheme for UAV-assisted data collection is proposed. This scheme decomposes the original problem into two sub-problems and solves them sequentially.For the node selection sub-problem, a multi-objective optimization model is proposed that integrates an energy-balancing mechanism with the objective of minimizing UAV flight distance. This model explicitly incorporates an upper bound on CS reconstruction error as a hard constraint, thereby jointly optimizing sensing performance and system energy efficiency while ensuring the recoverability of high-accuracy, full-network sensing data. This design effectively prevents premature failure of individual nodes caused by frequent activation, thereby enhancing the long-term operational stability of the network. Building upon this formulation, a Compressive Sensing-based Node Selection (CSNS) algorithm is developed to efficiently solve the multi-objective problem, yielding an optimal set of observed nodes and the corresponding CS measurement matrix that meet the system-level energy optimization requirements.For the energy consumption iterative optimization sub-problem, a two-stage joint optimization (TSJO) algorithm is proposed to address the non-convex optimization problem. Firstly, the UAV flight trajectory is fixed to search for the optimal UAV flight duration and SNs scheduling strategy in stage 1. Subsequently, the optimal UAV trajectory is solved in stage 2 based on the optimization results of stage 1. Finally, by alternately iterating between the two aforementioned sub-problems, the algorithm progressively approaches a high-quality suboptimal solution to the original mixed-integer nonlinear problem until convergence is achieved. This enables the full exploitation of the UAV’s mobility advantages to further unlock the system’s energy efficiency potential, all while ensuring the prescribed sensing performance.The rest of this paper is arranged as follows. “System model and problem formulation” section establishes an elaborate explanation of the system model and formulates the energy consumption minimization problem. “An efficient scheme to solve problem P” section outlines the energy consumption optimization scheme for UAV-assisted data collection by adopting the CS technique and introduces the specific processing of the CSNS algorithm and TSJO algorithm, respectively. “Performance evaluations” section conducts comprehensive simulations to validate the proposed approach, featuring comparative analysis with benchmark methods. Finally, “Conclusion” section concludes the paper by summarizing the key findings and contributions of this research. To enhance the representation of the system model, key system parameters and variables are listed in Table 1.Table 1Key system model notations.NotationMeaningNotationMeaningK, M, NNumber of SNs, the selected SNs and discrete path segments\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\boldsymbol{w}_k$$\end{document}Position coordinates of the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{SN}_{k}}$$\end{document}\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\boldsymbol{\Phi , \tilde{\Phi }, \Psi }$$\end{document}Observation matrix, path matrix and sparse basis\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$D(k), U_r(k)$$\end{document}Distance cost objective function and access cost objective of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{SN}_{k}}$$\end{document}\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\boldsymbol{z}, \boldsymbol{s}, \boldsymbol{y}$$\end{document}Sensing data vector, sparse signal vector and observation data vector\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$p^k_{\textrm{LoS}}$$\end{document}Probability of the LoS link between UAV and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{SN}_{k}}$$\end{document}\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$P_k, P_{\textrm{uav}}$$\end{document}Power of the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{SN}_{k}}$$\end{document} and the UAV\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$h_k(n)$$\end{document}Communication channel gain of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{SN}_{k}}$$\end{document} at the n-th path segment\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$E_s, E_{\textrm{uav}}$$\end{document}Energy consumption of SNs and UAV\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\theta$$\end{document}Maximum communication coverage angle of UAV\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$a_k(n)$$\end{document}Binary variable indicating whether \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{SN}_{k}}$$\end{document} is scheduled to transmit to the UAV at the n-th path segment\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\omega$$\end{document}Weighting factor balancing UAV and SNs energy consumptionT(n)UAV flight duration at the n-th path segment\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\lambda$$\end{document}Energy balance factor between the UAV and SNs\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\boldsymbol{q}(n)$$\end{document}Position coordinates of the UAV at the n-th path segmentf(k)Cost function of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{SN}_{k}}$$\end{document} By deeply integrating signal processing theory with intelligent optimization methods, this work formulates an energy minimization problem model for the UAV–SNs system based on sparse compressive sensing (CS)-driven data collection. To overcome the difficulties arising from the non-convex problem and the absence of closed-form expressions, an energy optimization scheme for UAV-assisted data collection is proposed. This scheme decomposes the original problem into two sub-problems and solves them sequentially. By deeply integrating signal processing theory with intelligent optimization methods, this work formulates an energy minimization problem model for the UAV–SNs system based on sparse compressive sensing (CS)-driven data collection. To overcome the difficulties arising from the non-convex problem and the absence of closed-form expressions, an energy optimization scheme for UAV-assisted data collection is proposed. This scheme decomposes the original problem into two sub-problems and solves them sequentially. For the node selection sub-problem, a multi-objective optimization model is proposed that integrates an energy-balancing mechanism with the objective of minimizing UAV flight distance. This model explicitly incorporates an upper bound on CS reconstruction error as a hard constraint, thereby jointly optimizing sensing performance and system energy efficiency while ensuring the recoverability of high-accuracy, full-network sensing data. This design effectively prevents premature failure of individual nodes caused by frequent activation, thereby enhancing the long-term operational stability of the network. Building upon this formulation, a Compressive Sensing-based Node Selection (CSNS) algorithm is developed to efficiently solve the multi-objective problem, yielding an optimal set of observed nodes and the corresponding CS measurement matrix that meet the system-level energy optimization requirements. For the node selection sub-problem, a multi-objective optimization model is proposed that integrates an energy-balancing mechanism with the objective of minimizing UAV flight distance. This model explicitly incorporates an upper bound on CS reconstruction error as a hard constraint, thereby jointly optimizing sensing performance and system energy efficiency while ensuring the recoverability of high-accuracy, full-network sensing data. This design effectively prevents premature failure of individual nodes caused by frequent activation, thereby enhancing the long-term operational stability of the network. Building upon this formulation, a Compressive Sensing-based Node Selection (CSNS) algorithm is developed to efficiently solve the multi-objective problem, yielding an optimal set of observed nodes and the corresponding CS measurement matrix that meet the system-level energy optimization requirements. For the energy consumption iterative optimization sub-problem, a two-stage joint optimization (TSJO) algorithm is proposed to address the non-convex optimization problem. Firstly, the UAV flight trajectory is fixed to search for the optimal UAV flight duration and SNs scheduling strategy in stage 1. Subsequently, the optimal UAV trajectory is solved in stage 2 based on the optimization results of stage 1. Finally, by alternately iterating between the two aforementioned sub-problems, the algorithm progressively approaches a high-quality suboptimal solution to the original mixed-integer nonlinear problem until convergence is achieved. This enables the full exploitation of the UAV’s mobility advantages to further unlock the system’s energy efficiency potential, all while ensuring the prescribed sensing performance. For the energy consumption iterative optimization sub-problem, a two-stage joint optimization (TSJO) algorithm is proposed to address the non-convex optimization problem. Firstly, the UAV flight trajectory is fixed to search for the optimal UAV flight duration and SNs scheduling strategy in stage 1. Subsequently, the optimal UAV trajectory is solved in stage 2 based on the optimization results of stage 1. Finally, by alternately iterating between the two aforementioned sub-problems, the algorithm progressively approaches a high-quality suboptimal solution to the original mixed-integer nonlinear problem until convergence is achieved. This enables the full exploitation of the UAV’s mobility advantages to further unlock the system’s energy efficiency potential, all while ensuring the prescribed sensing performance. Table 1Key system model notations.NotationMeaningNotationMeaningK, M, NNumber of SNs, the selected SNs and discrete path segments\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\boldsymbol{w}_k$$\end{document}Position coordinates of the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{SN}_{k}}$$\end{document}\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\boldsymbol{\Phi , \tilde{\Phi }, \Psi }$$\end{document}Observation matrix, path matrix and sparse basis\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$D(k), U_r(k)$$\end{document}Distance cost objective function and access cost objective of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{SN}_{k}}$$\end{document}\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\boldsymbol{z}, \boldsymbol{s}, \boldsymbol{y}$$\end{document}Sensing data vector, sparse signal vector and observation data vector\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$p^k_{\textrm{LoS}}$$\end{document}Probability of the LoS link between UAV and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{SN}_{k}}$$\end{document}\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$P_k, P_{\textrm{uav}}$$\end{document}Power of the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{SN}_{k}}$$\end{document} and the UAV\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$h_k(n)$$\end{document}Communication channel gain of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{SN}_{k}}$$\end{document} at the n-th path segment\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$E_s, E_{\textrm{uav}}$$\end{document}Energy consumption of SNs and UAV\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\theta$$\end{document}Maximum communication coverage angle of UAV\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$a_k(n)$$\end{document}Binary variable indicating whether \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{SN}_{k}}$$\end{document} is scheduled to transmit to the UAV at the n-th path segment\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\omega$$\end{document}Weighting factor balancing UAV and SNs energy consumptionT(n)UAV flight duration at the n-th path segment\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\lambda$$\end{document}Energy balance factor between the UAV and SNs\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\boldsymbol{q}(n)$$\end{document}Position coordinates of the UAV at the n-th path segmentf(k)Cost function of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{SN}_{k}}$$\end{document} Key system model notations. System model and problem formulation CS-based data collection model The data collection network studied in this paper is deployed in remote areas that lacks ground mobile base stations, and its network topology consists of several terrestrial SNs, an airborne UAV mobile aggregation node, and a low-Earth-orbit (LEO) satellite network, as shown in Fig. 1. Assuming the system is aware of the locations of all SNs in the sensing area in advance. A rotary-wing UAV acts as a mobile aerial base station and periodically takes off from the starting point to perform flight missions with optimized flight trajectory, flight time and SNs scheduling strategy, and lands at the end point to wait for the next round of flight missions. As a mobile Sink, the UAV uses CS sparse sampling to gather sensory data of some SNs in the monitoring area during each round of data collection, and transmits the gathered data to the remote server via low-orbit satellites. The server employs the CS reconstruction algorithm to reconstruct sensory data of entire network.Fig. 1System model of UAV-assisted WSNs for data collection. System model of UAV-assisted WSNs for data collection. Assume that K SNs are randomly deployed in the monitoring region of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$L\times L$$\end{document}. The horizontal position of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\textit{SN}_{k}$$\end{document} is denoted as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\boldsymbol{w}}_{\boldsymbol{k}}}=\left( {{x}_{k}},{{y}_{k}} \right)$$\end{document} and the sensing data vector of the SNs in the whole network is represented by \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\boldsymbol{z}={{\left( {{z}_{1}},{{z}_{2}},\ldots ,{{z}_{K}} \right) }^{T}}$$\end{document}. Due to the spatial correlation between the sensory data, the vector \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\boldsymbol{z}$$\end{document} can be expressed as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\boldsymbol{z}=\boldsymbol{\Psi }\boldsymbol{s}$$\end{document}, where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\boldsymbol{\Psi } \in {{\mathbb {R}}^{K\times K}}$$\end{document} is a sparse basis, and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\boldsymbol{s}$$\end{document} is the sparse signal after the sparse transformation of data vector \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\boldsymbol{z}$$\end{document}. Here, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\boldsymbol{s}}}$$\end{document} contains only \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$s\left( s\ll K \right)$$\end{document} non-zero elements26. Then the data vector \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\boldsymbol{y}\in {{\mathbb {R}}^{K\times 1}}$$\end{document} collected by the UAV can be denoted as:1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \boldsymbol{y}=\boldsymbol{\Phi } \boldsymbol{z} + \boldsymbol{\chi }=\boldsymbol{\Phi \Psi \boldsymbol{s}}+ \boldsymbol{\chi } \end{aligned}$$\end{document}where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\boldsymbol{\Phi }\,\in {{\mathbb {R}}^{M\times K}}$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left( M\ll K \right)$$\end{document} is the observation matrix, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\boldsymbol{\chi } \in {{\mathbb {R}}^{K}}$$\end{document} represents the inevitable system noise. CS theory states that if the restricted isometry property (RIP) is satisfied by the observation matrix and sparse basis27, the sparse signal \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\boldsymbol{s}$$\end{document} can be successfully reconstructed with overwhelming probability by solving a convex optimization problem of Eq. (2).2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \underset{\boldsymbol{s}}{{\min }}\,{{\left\| \boldsymbol{s} \right\| _{1}}} \quad s.t.\quad \left\| \boldsymbol{y}-\boldsymbol{\Phi \Psi }\boldsymbol{\hat{s}} \right\| _{2}^{2}\le \varepsilon \end{aligned}$$\end{document}where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\boldsymbol{\hat{s}}$$\end{document} represents the reconstructed sparse signal, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\left\| \boldsymbol{s} \right\| _{1}}}$$\end{document} denotes the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${L}_{1}$$\end{document} norm of the vector \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\boldsymbol{s}$$\end{document}, which is defined as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\left\| \boldsymbol{s} \right\| _{1}}}=\sum _{i=1}^{K}|{\textit{s}_i|}$$\end{document}. Since the selection of different CS sampling nodes will result in varying reconstruction errors, we introduce the constraint C1 to prevent the CS reconstruction error from exceeding the maximum threshold \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\varepsilon$$\end{document} allowed by the system during the data collection process, where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$G\left( \cdot \right)$$\end{document} denotes the reconstruction operation of the sparse signal.3\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \ C1{:}\left\| G(\boldsymbol{y},\boldsymbol{\Phi },\boldsymbol{\Psi })-\boldsymbol{s}\right\| _{2}^{2}\le \varepsilon \end{aligned}$$\end{document} Energy consumption model The system energy consumption during the data collection process primarily encompasses the energy consumption of UAV and SNs in this work. Concerning UAV energy consumption, UAV flight energy is the dominant energy consumption, while communication and computing energy consumption are usually negligible12. Similar to8, UAV flight power can be expressed as a function of flight speed \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$v(t)$$\end{document}, where t is a continuous time variable. The UAV flight energy consumption can be expressed as:4\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \ E_{uav}=\int _{0}^{T}P_{uav}(v(t))dt \end{aligned}$$\end{document}where T is the total UAV flight duration and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$P_{uav}(v(t))$$\end{document} denotes the UAV flight power. The UAV trajectory can be described as a time-varying two-dimensional horizontal coordinate \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\boldsymbol{q(t)}=\left( {{x}_{u}}\left( t \right) ,{{y}_{u}}\left( t \right) \right)$$\end{document}, with a fixed flight altitude H. The temporal continuity of UAV trajectory makes the trajectory design problem unsolvable, most existing studies adopted a time discretization approach to divide UAV trajectory into \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$N$$\end{document} time segments. However, this approach requires the total UAV flight duration to be determined in advance, which contradicts the goal of UAV energy optimization in this paper. Hence, we adopt the UAV trajectory discretization approach to decompose the UAV trajectory. Let the UAV trajectory be decomposed into N discrete path segments, then the corresponding N+1 trajectory points are denoted as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left\{ \boldsymbol{q(n)} \right\} _{n=0}^{N}$$\end{document}, which \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\boldsymbol{q(0)} ={{\boldsymbol{q}}_{\boldsymbol{I}}}$$\end{document}, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\boldsymbol{q(N)}={{\boldsymbol{q}}_{\boldsymbol{F}}}$$\end{document} denote the initial and final points of UAV, respectively. At the nth path segment, UAV flight duration can be represented as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\textit{T}\left( n \right)$$\end{document}, and the total UAV flight duration is calculated as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$T=\sum _{n=1}^{N}T(n)$$\end{document}. Then UAV trajectory constraints can be written as follows:5\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \ C2:{\left\{ \begin{array}{ll}\boldsymbol{q(0)} ={{\boldsymbol{q}}_{\boldsymbol{I}}},\boldsymbol{q(N)}={{\boldsymbol{q}}_{\boldsymbol{F}}}\\ \parallel \boldsymbol{q(n)}-\boldsymbol{q(n-1)}\parallel \le \min \{\Delta _{\max },V_{\max }T(n)\},& \forall n\end{array}\right. } \end{aligned}$$\end{document}where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{V}_{\max }}$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\!\!\Delta \!\!}_{\max }}$$\end{document} are maximum flight speed and the maximum flight distance allowed by UAV in each segment, respectively. Limiting the length of each discrete path segment to be sufficiently short, UAV flight speed \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$v\left( n \right)$$\end{document} within a single path segment can be considered as a fixed value and denoted as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$v\left( n \right) ={\parallel \boldsymbol{q(n)}-\boldsymbol{q(n-1)}\parallel }/{T\left( n \right) }\;={\Delta \left( n \right) }/{T\left( n \right) }\;$$\end{document}. After discretizing the UAV trajectory, the UAV energy consumption can be rewritten as:6\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \ {{E}_{uav}}=\sum \limits _{n=1}^{N}{{{P}_{uav}}\left( v\left( n \right) \right) T\left( n \right) } \end{aligned}$$\end{document}where the mathematical model of UAV flight power is given as:7\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \ \begin{aligned}{P}_{uav}\big (v\big (n\big )\big )=&P_{0}\Bigg (1+\frac{3v^{2}\left( n\right) }{U_{tip}^{2}}\Bigg )+P_{1}\sqrt{\sqrt{1+\frac{v^{4}\left( n\right) }{4v_{0}^{4}}}-\frac{v^{2}\left( n\right) }{2v_{0}^{2}}}+\frac{1}{2}d_{0}\rho uAv^{3}\left( n\right) \end{aligned} \end{aligned}$$\end{document}In UAV hovering status, the UAV power can be denoted by where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{P}_{0}}+{{P}_{1}}$$\end{document}. The parameters \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$U_{tip}^{{}}$$\end{document}, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{d}_{0}}$$\end{document}, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\rho$$\end{document}, A, u and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$v_{0}^{{}}$$\end{document} are constants related to UAV equipment and environment11. Concerning SNs energy consumption, we focus solely on their communication energy consumption. SNs employ a sleep-wake scheduling mechanism during the data collection process. We define a binary access scheduling matrix \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$A={{\left( {{a}_{k}}\left( n \right) \right) }_{K\times N}}$$\end{document}, where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{a}_{k}}\left( n \right) =1$$\end{document} implies that \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{SN}_{k}}$$\end{document} establishes a transmission link with UAV in the nth path segment, otherwise, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{a}_{k}}\left( n \right) =0$$\end{document}. Assuming SNs transmit data at a fixed communication power \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{P}_{k}}$$\end{document}, the SNs energy consumption is modeled as:8\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \ E_{s}=P_{k}\sum _{n=1}^{N}\sum _{k=1}^{K}a_{k}(n)T(n) \end{aligned}$$\end{document} Communication model Assuming that UAV communication module works in duplex mode, UAV can forward the sensed data to the LEO satellite side at the same time after receiving it. Usually, the channel capacity of the air-space link is greater than that of the air-ground link. The air-space link communication can be simplified to real-time transparent communication. Therefore, this paper mainly analyzes the air-ground link communication process. Due to the absence of traditional mobile base station networks or other network coverage in remote areas, communication interference in the wireless channel primarily stems from natural environmental factors and other non-human-related sources. Therefore, we describe the air-to-ground communication link employing a probabilistic line-of-sight (LoS) model16. The probability that a LoS link exists between UAV and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$SN_{k}$$\end{document} can be calculated as:9\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \ p_{LoS}^{k}=\frac{1}{1+c_{1}\textrm{exp}\Big [-c_{2}\Big (\theta _{k}(n)-c_{1}\Big )\Big ]} \end{aligned}$$\end{document}where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$c_{1}$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$c_{2}$$\end{document} are constants related to environment variables, and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\theta }_{k}}\left( n \right)$$\end{document} is the angle of elevation between \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$SN_{k}$$\end{document} and UAV, which is denoted as:10\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \ \theta _{k}\left( n\right) =\frac{180}{\pi }\arctan \left( \frac{H}{\left\| \boldsymbol{w}_{k}-{{\boldsymbol{q}}}\left( n\right) \right\| ^{2}}\right) \end{aligned}$$\end{document}The probability of the existence of an NLoS link is calculated as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$p_{NLoS}^{k}=1-p_{LoS}^{k}$$\end{document}. Due to the probabilistic randomness of LoS and NLoS links, the air-ground link channel gain is a random variable. By averaging two random variables, the expected channel gain is calculated as:11\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} h_{k}\left( n\right) =\frac{p_{LoS}^{k}h_{0}}{d_{k}^{2}\left( n\right) }+\frac{\left( 1-p_{LoS}^{k}\right) \beta h_{0}}{d_{k}^{2}\left( n\right) }=\frac{\tilde{P}_{LoS}^{k}h_{0}}{\left\| \boldsymbol{w}_{k}-{{\boldsymbol{q}}}(n)\right\| ^{2}+H^{2}} \end{aligned}$$\end{document}where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\beta <1$$\end{document} is the fading coefficient of the NLoS link, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\tilde{P}_{L\text {o}S}^{k}=p_{LoS}^{k}+\left( 1-p_{LoS}^{k} \right) \beta$$\end{document} represents the normalized LoS probability incorporating the NLoS channel attenuation effect, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{d}_{k}}\left( n \right)$$\end{document} is the distance between \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$SN_{k}$$\end{document} and UAV, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{h}_{0}}$$\end{document} is the reference gain at 1 m and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left\| \cdot \right\|$$\end{document} denotes the Euclidean paradigm. It can be observed that the regularized LoS probability \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\tilde{P}_{L\text {o}S}^{k}$$\end{document} is correlated with the UAV trajectory \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\boldsymbol{{q}}}}\left( {{\boldsymbol{n}}} \right)$$\end{document}, which makes the UAV trajectory optimization problem difficult to solve. Since different UAV trajectories will affect the UAV communication elevation angle and have quite limited impact on the communication rate11, for simplification in analysis, we approximate the regularized LoS probability \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\tilde{P}_{L\text {o}S}^{k}$$\end{document} adopting the mean approximation method, that is, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\tilde{P}_{L\text {o}S}^{k}\approx \bar{P}_{L\text {o}S}^{k}$$\end{document}, where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\bar{P}_{L\text {o}S}^{k}$$\end{document} denotes the regularized LoS probability obtained by adopting the average communication elevation angle. Further, the communication rate between \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$SN_{k}$$\end{document} and UAV can be expressed as:12\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} R_{k}\left( n\right) =B\log _{2}\left( 1+\frac{\gamma _{k}}{\left\| \boldsymbol{w}_{k}-{{\boldsymbol{q}}}(n)\right\| ^{2}+H^{2}}\right) \end{aligned}$$\end{document}where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\gamma }_{k}}={\left( {{P}_{k}}\bar{P}_{L\text {o}S}^{k}{{h}_{0}} \right) }/{{{N}_{0}}}\;$$\end{document}, B represents the channel bandwidth and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{N}_{0}}$$\end{document} represents the noise power. During the UAV-assisted data collection process, in order to avoid co-channel signal interference, the time division multiple access (TDMA) technique is utilized for data transmission. UAV communicates with at most one node within any given path segment. Then there exists the following constraint:13\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} C3:{\left\{ \begin{array}{ll}a_{k}(n)\in \{0,1\},& \forall k,\forall n\\ \sum _{k=1}^{K}a_{k}(n)\le 1,& \forall n\\ \end{array}\right. } \end{aligned}$$\end{document}Define the set of SNs participating in sparse CS sampling as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\kappa =\left\{ S{{N}_{k}}\left| {{a}_{k}}\left( n \right) =1,\forall n \right. \right\}$$\end{document}. To guarantee the complete transmission of each SN’s data, we constrain the transmission task value of any SN in the set \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\kappa$$\end{document} to be no less than \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{Q}_{K}}$$\end{document}, which is written as:14\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} C4: \sum _{n=1}^{N}a_{k}(n)T(n)R_{k}(n)\ge Q_{K},\quad \forall k\in \kappa \end{aligned}$$\end{document}Furthermore, since the performance of CS-based data collection algorithms is sensitive to packet loss in communication links28, the constraint C5 is imposed on the communication distance between \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$SN_{k}$$\end{document} and UAV to ensure the reliability of air-to-ground data transmission, where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\theta$$\end{document} represents the maximum communication coverage angle of UAV.15\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} C5:a_{k}(n)\Vert \boldsymbol{w}_{k}-\boldsymbol{q}(n)\Vert \le H\tan \theta ,\quad \forall k,n \end{aligned}$$\end{document} Problem formulation Utilizing the efficient reconstruction capability of the CS technique, we apply the sparse CS technique to the UAV-assisted data collection process. By collecting data from only a subset of SNs to reconstruct data for the entire network, the flight distance of the UAV is significantly reduced, which in turn decreases the energy consumption of the UAV-SNs. Consequently, for each data collection round in WSNs, the energy consumption optimization problem of the UAV-SNs system can be formulated as:16\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \begin{aligned}&P:\min E\big \{{{\boldsymbol{A,Q,T}}}\big \}=\min _{\{{{\boldsymbol{A,Q,T}}}\}}\Big [\big (1-\omega \big )\lambda E_{s}+\omega E_{uav}\Big ]\\ &\; st. \ C1,C2,C3,C4,C5\end{aligned} \end{aligned}$$\end{document}where the optimization variables include UAV flight duration \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathbf {{{\boldsymbol{T}}}}=\left\{ T\left( n \right) ,\ \forall n \right\}$$\end{document}, UAV flight trajectory \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\boldsymbol{{\textbf {Q}}}}}=\left\{ \mathbf {{{\boldsymbol{q}}}}\left( n \right) ,\ \forall n \right\}$$\end{document}, and SNs scheduling strategy \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathbf {{{\boldsymbol{A}}}}=\left\{ {{a}_{k}}\left( n \right) ,\ \forall n,k \right\}$$\end{document}. The parameter \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\lambda$$\end{document} is the energy balance factor, which is used to adjust the deviation caused by the condition that the energy consumption of the UAV differs significantly in magnitude compared to that of the SN. The parameter \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\omega \left( 0\le \omega \le 1 \right)$$\end{document} represents the weight coefficient that measures the importance of UAV and SNs energy consumption. Constraint C1 limits the upper bound of the CS data reconstruction error. Constraint C2 indicates that the UAV trajectory starts from the initial point, lands to the final point, and also limits the maximum flight distance of each path segment. Constraint C3 ensures that at most one SN transmits data to the UAV at each path segment. Constraint C4 limits the minimum transmission task value for each selected node. Constraint C5 ensures that the SNR of the air-to-ground link exceeds the minimum threshold allowed by the system. An efficient scheme to solve problem P The energy consumption minimization problem P is extremely difficult to solve since the CS reconstruction accuracy and optimization variables in constraint C1 cannot be established in a mathematical closed-form expression, as well as the non-convexity of the constraint C4 and the objective function. A comprehensive analysis of problem P reveals that the lack of mathematical closed-form formulation arises from the challenge associated with integrating the CS technique with UAV-assisted data collection technique. If a well-designed node selection strategy for CS sparse sampling is developed, which satisfies the requirements of CS data reconstruction accuracy while simultaneously addressing UAV flight energy consumption and network energy balance, the aforementioned integration challenges can be effectively resolved. Therefore, an energy consumption optimization scheme for UAV-assisted data collection based on sparse CS technique is proposed in this paper, which decouples the original problem P into two sub-problems. Firstly, to address the lack of mathematical closed-form expression between CS reconstruction accuracy and optimization strategies, the node selection sub-problem \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S_{1}P$$\end{document} is defined, and the objectives of network energy balance and UAV flight distance minimization under the constraint of CS reconstruction accuracy can be realized by optimizing the node selection strategy. Secondly, relying on the node selection results of sub-problem \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S_{1}P$$\end{document}, the original problem P is transformed into the energy consumption iterative optimization sub-problem \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S_{2}P$$\end{document}. By solving the sub-problem \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S_{2}P$$\end{document}, the UAV-SNs energy consumption minimization problem P can be solved approximately. The overall solution to problem P implemented in this paper is illustrated in Fig. 2.Fig. 2The overall framework for solving problem P. The overall framework for solving problem P. Node selection strategy optimization According to the sparse CS theory29, different node selection strategies require the design of corresponding CS sparse measurement matrices, which in turn affect the precision of CS data reconstruction. Moreover, node selection strategies also influence the energy consumption balance of sensor network and the UAV flight distance. Therefore, the node selection subproblem \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S_{1}P$$\end{document} is a multi-objective optimization problem. For the objective of network energy balance, the access cost objective function for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$SN_{k}$$\end{document} is defined as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{U}_{r}}\left( k \right) ={{{J}_{r}}\left( k \right) }/{J_{r}^{\max }}\;$$\end{document}, where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{J}_{r}}\left( k \right)$$\end{document} represents the number of times \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$SN_{k}$$\end{document} has been selected until the rth round of data collection, and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$J_{r}^{\max }$$\end{document} denotes the maximum value of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{J}_{r}}\left( k \right)$$\end{document} in the rth round. Selecting \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$SN_{k}$$\end{document} with smaller values of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{U}_{r}}\left( k \right)$$\end{document} will encourage SNs with fewer participation times to participate, thereby making the network energy consumption more balanced. For the objective of UAV flight distance, the distance between \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$SN_{k}$$\end{document} and the line connecting the initial point \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\mathbf {{{\boldsymbol{q}}}}}_{\textbf{I}}}$$\end{document} and the final points \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{\boldsymbol{{\textbf {q}}}}}}_{\textbf{F}}}$$\end{document} is defined as:17\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} d\left( k \right) =\frac{\left| \left( {{\boldsymbol{w}}_{k}}-{{\boldsymbol{q}}_{\boldsymbol{I}}} \right) \times \left( {{\boldsymbol{q}}_{\boldsymbol{F}}}-{{\boldsymbol{q}}_{\boldsymbol{I}}} \right) \right| }{\left\| {{\boldsymbol{q}}_{\boldsymbol{F}}}-{{\boldsymbol{q}}_{\boldsymbol{I}}} \right\| } \end{aligned}$$\end{document}Then the distance cost objective function for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$SN_{k}$$\end{document} is defined as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$D\left( k \right) ={d\left( k \right) }/{{{d}_{\max }}}\;$$\end{document}, where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{d}_{\max }}$$\end{document} is the maximun value of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$d\left( k \right)$$\end{document}. Selecting \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$SN_{k}$$\end{document} with smaller values of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$D\left( k \right)$$\end{document} will result in a shorter UAV flight distance. To describe the cost of each selected SN regarding the aforementioned two optimization objectives, the cost function of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$SN_{k}$$\end{document} is defined using the classical linear weighting method as:18\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} f(k)=\omega _{0}U_{r}(k)+(1-\omega _{0})D(k) \end{aligned}$$\end{document}where the factor \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$0\le {{\omega }_{0}}\le 1$$\end{document} serves as a weight to balance the node access rate and the cost of node distance. The access cost objective function \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{U}_{r}}\left( k \right)$$\end{document} is related to the historical access status of each SN, so the cost function \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$f(k)$$\end{document} will be updated successively during each round of data collection. The above process provides a cost evaluation model for node selection. Next, we will discuss the constraint on the CS reconstruction accuracy. The set of selected SNs is denoted as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\kappa =\left\{ S{{N}_{{{k}_{1}}}},S{{N}_{{{k}_{2}}}},\cdots ,S{{N}_{{{k}_{M}}}} \right\}$$\end{document}, where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{k}_{i}}$$\end{document} denotes the ID number of the selected node. Then the CS observation matrix is designed as:19\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \boldsymbol{\Phi }=\left( \phi _{mk}\right) _{M\times K}={\left\{ \begin{array}{ll}1,& m=i,k=k_{i}\\ 0,& otherwise\end{array}\right. } \end{aligned}$$\end{document}Given the tight coupling between the node selection strategy and the UAV flight trajectory, we define the path matrix \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\boldsymbol{\tilde{\Phi }}\in {{\mathbb {R}}^{K\times K}}$$\end{document}, which is a diagonal matrix with the following definition for the diagonal elements:20\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \tilde{\phi }_{k,k}={\left\{ \begin{array}{ll}1,& k=k_{i},\\ 0,& otherwise\end{array}\right. } \end{aligned}$$\end{document}According to matrix transformation theory, the relationship between the path matrix \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\boldsymbol{\tilde{\Phi }}$$\end{document} and the observation matrix \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\boldsymbol{\Phi }\,$$\end{document} can be expressed as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\boldsymbol{\tilde{\Phi }}=\boldsymbol{\Phi }{{\,}^{T}}\boldsymbol{\Phi }\,$$\end{document}. To ensure adherence to constraint C1, we introduce the mean squared error (MSE) of least squares estimation as a metric for evaluating the CS reconstruction accuracy30. The least squares estimate of the reconstructed signal is written as:21\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \hat{\boldsymbol{s}}=\boldsymbol{\left( \Psi ^{T}\Phi ^{T}\Phi \Psi \right) ^{-1}\Psi ^{T}\Phi ^{T}\boldsymbol{y}=\left( \Psi ^{T}\tilde{\Phi }\Psi \right) ^{-1}\Psi ^{T}\tilde{\Phi }\Psi }\boldsymbol{s} \end{aligned}$$\end{document}Then the MSE of the least squares estimate of the reconstructed signal is given by:22\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} E\Big (\Big \Vert \boldsymbol{s}-\hat{\boldsymbol{s}}\Big \Vert ^{2}\Big )=\sigma ^{2}\operatorname {Tr}\Big (\Big (\boldsymbol{\Psi ^{T}\tilde{\Phi }\Psi \Big )^{-1}\Big )} \end{aligned}$$\end{document}where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\sigma }^{2}}$$\end{document} is variance and the trace of a matrix is denoted by \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\operatorname {Tr}(\cdot )$$\end{document}. Therefore, constraint C1 can be reformulated as a threshold constraint based on the MSE. Given that the data reconstruction performance of Gaussian random observation matrices has been repeatedly validated, the MSE threshold constraint can be set as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\eta =\operatorname {Tr}\left( {{\left( {{\boldsymbol{\Psi }}^{\boldsymbol{T}}}{{{\boldsymbol{\tilde{\Phi }}}}^{rand}}\boldsymbol{\Psi } \right) }^{-1}} \right)$$\end{document}, where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\boldsymbol{\tilde{\Phi }}}^{rand}}$$\end{document} represents the path matrix corresponding to Gaussian random node selection strategy. Thus, constraint C1 can be replaced by Eq. (23).23\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \operatorname {Tr}\biggl (\left( \boldsymbol{\Psi ^{T}\tilde{\Phi }\Psi }\right) ^{-1}\biggr )\le \eta \end{aligned}$$\end{document}Through the above modeling, the node selection sub-problem can be converted into a tractable form \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S_{1}P_{1}$$\end{document}, which is to select M SNs from K SNs and minimize the sum of the costs of the M SNs under the condition of satisfying the CS reconstruction accuracy constraint, as shown in Eq. (). 24a\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} S_{1}P_{1}: \min _{\tilde{\boldsymbol{\Phi }}} \quad&\sum _{k=1}^{K} \tilde{\phi }_{k,k} f(k) \end{aligned}$$\end{document}24b\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} s.t. \quad&\tilde{\phi }_{k,k} \in \{0,1\}, \quad k = 1,...,K \end{aligned}$$\end{document}24c\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned}&\operatorname {Tr}\big (\tilde{\boldsymbol{\Phi }}\big )=M \end{aligned}$$\end{document}24d\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned}&\operatorname {Tr}\biggl (\left( \boldsymbol{\Psi ^{T}\tilde{\Phi }\Psi }\right) ^{-1}\biggr )\le \eta \end{aligned}$$\end{document} where the constraints Eq. (24b) and Eq. (24c) limit the number of nodes that can participate in data collection to M, and constraint Eq. (24d) limits the reconstruction error to no more than the maximum allowable threshold. For constraint Eq. (24d), \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\boldsymbol{\tilde{\Phi }}$$\end{document} is a diagonal matrix and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\tilde{\phi }}_{k,k}}$$\end{document} is non-negative, thus \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\boldsymbol{\Psi }}^{\boldsymbol{T}}}\boldsymbol{\tilde{\Phi }\Psi }$$\end{document} is a positive definite symmetric matrix. According to the theory of31, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\operatorname {Tr}\left( {{\boldsymbol{X}}^{-1}} \right)$$\end{document} is convex when \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\boldsymbol{X}$$\end{document} is a positive definite symmetric matrix, so constraint Eq. (24d) is a convex constraint. However, problem \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S_{1}P_{1}$$\end{document} is still non-convex due to the binary nature of the constraint Eq. (24b). To solve the above non-convex problem, the binary variable is relaxed to a continuous variable ranging from 0 to 1. Then \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S_{1}P_{1}$$\end{document} is transformed into the following optimization problem:25\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \begin{aligned} S_{1}P_{2}:\min _{\hat{\boldsymbol{\Phi }}}\sum _{k=1}^{K}\hat{\phi }_{k,k}f(k) \\ s.t.\; (24c),(24d) \; \;\\ 0\le \hat{\phi }_{k,k}\le 1,\forall k \end{aligned} \end{aligned}$$\end{document}By employing classical convex optimization algorithms, the problem \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S_{1}P_{2}$$\end{document} can be effectively solved, yielding an optimized path matrix \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\boldsymbol{\hat{\Phi }}$$\end{document}. It should be noted that the values \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\hat{\phi }}_{k,k}}$$\end{document} in the optimal path matrix are not necessarily binary integers. Following the heuristic principle, we select the top M values of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\hat{\phi }}_{k,k}}$$\end{document} and adopt their corresponding index as the selected node ID. The specific procedure is concluded in Algorithm 1. It should be noted that since only some SNs participate in each data collection round, the SNs scheduling strategy needs to be updated successively according to the historical participation of the SNs to ensure balanced energy consumption in the sensor network. That is to say, the output results of Algorithm 1 are different in each data collection round. Algorithm 1CS-based node selection algorithm. CS-based node selection algorithm. Energy consumption iterative optimization By solving the node selection sub-problem \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S_{1}P$$\end{document}, constraint C1 of the original problem P has been successfully fulfilled. Utilizing the obtained node selection strategy, the original problem P is converted into the energy iteration sub-problem \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S_{2}P_{1}$$\end{document}, which is mathematically modeled as following.26\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \begin{aligned}&{{S}_{2}}{{P}_{1}}:\underset{\left\{ \boldsymbol{{{\boldsymbol{A}}}},\boldsymbol{{{\boldsymbol{Q}}}},\boldsymbol{{{\boldsymbol{T}}}} \right\} }{{\min }}\,\left\{ \omega \sum \limits _{n=1}^{N}{\left[ {{P}_{0}}\left( T\left( n \right) +\frac{3{{\Delta }^{2}}\left( n \right) }{U_{tip}^{2}T\left( n \right) } \right) +{{P}_{1}}{{\left( \sqrt{{{{T}^{4}}\left( n \right) }+\frac{{{\Delta }^{4}}\left( n \right) }{4v_{0}^{4}}}-\frac{{{\Delta }^{2}}\left( n \right) }{2v_{0}^{2}} \right) }^{\frac{1}{2}}}+\frac{1}{2}{{d}_{0}}\rho uA\frac{{{\Delta }^{3}}\left( n \right) }{{{T}^{2}}\left( n \right) } \right] } \right. \\&\quad \quad \, \,\,\quad \quad \,\,\quad \left. +\lambda \left( 1-\omega \right) \sum \limits _{n=1}^{N}{\sum \limits _{k=1}^{K}{{{a}_{k}}\left( n \right) {{P}_{k}}T\left( n \right) }} \right\} \\&\quad \quad \, \,\,\quad s.t.\ \ C2-C5 \\ \end{aligned} \end{aligned}$$\end{document}Since the optimization objective function in \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S_{2}P_{1}$$\end{document} is a non-convex function concerning the optimization variables \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left\{ \boldsymbol{{{\boldsymbol{A}}}},\boldsymbol{{{\boldsymbol{Q}}}},\boldsymbol{{{\boldsymbol{T}}}} \right\}$$\end{document}, in order to convert it into a tractable convex function, a slack variable \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$z\left( n \right)$$\end{document} is introduced, whose mathematical expression is given by:28\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} z^{2}\left( n\right) \ge \sqrt{T^{4}\left( n\right) +\frac{\Delta ^{4}\left( n\right) }{4v_{0}^{4}}}-\frac{\Delta ^{2}\left( n\right) }{2v_{0}^{2}}, \forall n \end{aligned}$$\end{document}For ease of processing, Eq. (28) can be mathematically transformed into:29\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \frac{T^{4}(n)}{z^{2}(n)}\le z^{2}(n)+\frac{\Delta ^{2}(n)}{\nu _{0}^{2}},\forall n \end{aligned}$$\end{document}Then problem \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S_{2}P_{1}$$\end{document} is transformed into problem \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S_{2}P_{2}$$\end{document} shown in Eq. (30), where the new constraint Eq. (29) is incorporated to ensure that problem \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S_{2}P_{2}$$\end{document} is equivalent to \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S_{2}P_{1}$$\end{document}.30\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \begin{aligned}&{{S}_{2}}{{P}_{2}}:\underset{\left\{ \boldsymbol{A,Q,T,Z} \right\} }{ \text {min}}\,\left\{ {{E}_{sum}} \right\} \\&\quad \ s.t.\ \ C2-C5,(28) \\ \end{aligned} \end{aligned}$$\end{document}where,31\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} {{E}_{sum}}=\omega \sum \limits _{n=1}^{N}{\left[ {{P}_{0}}\left( T\left( n \right) +\frac{3{{\Delta }^{2}}\left( n \right) }{U_{tip}^{2}T\left( n \right) } \right) +{{P}_{1}}z\left( n \right) +\frac{1}{2}{{d}_{0}}\rho uA\frac{{{\Delta }^{3}}\left( n \right) }{{{T}^{2}}\left( n \right) } \right] }+\lambda \left( 1-\omega \right) \sum \limits _{n=1}^{N}{\sum \limits _{k=1}^{K}{{{a}_{k}}\left( n \right) {{P}_{k}}}}T\left( n \right) \end{aligned}$$\end{document}However, constraints C4 and Eq. (29) are non-convex constraints on the optimization variables and there exists a binary integer variable in the constraint C3. The problem \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S_{2}P_{2}$$\end{document} is a mixed-integer non-convex optimization problem. This work proposes a TSJO algorithm that leverages the block coordinate descent (BCD) technique to tackle the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S_{2}P_{2}$$\end{document}, thereby reducing its computational complexity. The algorithm decomposes the original problem into two stages and iteratively alternates between them to progressively approach the global optimum. This decomposition is grounded on the following foundations: 1) By discretizing the UAV trajectory into a sequence of waypoints and treating the flight duration of each segment as an independent variable, the spatial and temporal dimensions are formally decoupled; 2) An explicit maximum flight speed constraint is introduced to bidirectionally propagate dynamic feasibility information between the two stages, ensuring that the alternating optimization remains within the physically realizable domain throughout. Building upon the above modeling and decomposition strategy and supported by the convergence guarantee inherent to the BCD framework, the TSJO algorithm proceeds as follows: 1) Optimization of SNs Scheduling and UAV Flight Duration: Given the UAV trajectory \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\boldsymbol{{{\boldsymbol{Q}}}}$$\end{document}, the UAV flight duration \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\boldsymbol{T}$$\end{document} and SNs scheduling \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\boldsymbol{A}$$\end{document} are jointly optimized in stage 1. The problem \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S_{2}P_{2}$$\end{document} is reformulated as:32\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \begin{aligned}&{{S}_{2}}{{P}_{3}}:\underset{\left\{ \boldsymbol{A,T,Z} \right\} }{ \text {min}}\,\left\{ {{E}_{sum}} \right\} \\&\quad \quad \ s.t.\ \ C3-C5,(28) \\ \end{aligned} \end{aligned}$$\end{document}To transform this optimization issue into a convex form, the value of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{a}_{k}}\left( n \right)$$\end{document} is relaxed to a continuous value \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$0\le {{\hat{a}}_{k}}\left( n \right) \le 1$$\end{document}. At the same time, an auxiliary variable \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\tau }_{k}}\left( n \right) ={{\hat{a}}_{k}}\left( n \right) T\left( n \right)$$\end{document} is introduced, then C3 can be reformulated as:33\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} C7:{\left\{ \begin{array}{ll}0\le \hat{a}_{k}\left( n\right) \le 1,\;\forall k,\forall n\\ \sum \limits _{k=1}^{K}{{\hat{a}_{k}}\left( n \right) }\le 1,\; \forall n \\ \sum \limits _{k=1}^{K}{{{\tau }_{k}}\left( n \right) }\le T\left( n \right) ,\; \forall n \end{array}\right. } \end{aligned}$$\end{document}The constraint C5 is reformulated as:34\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} C8:{{\tau }_{k}}\left( n \right) \left\| {{\mathbf {{{\boldsymbol{w}}}}}_{k}}-{{\boldsymbol{{q}}}}\left( n \right) \right\| \le H\tan \theta T\left( n \right) , \ \ \forall k,\forall n \end{aligned}$$\end{document}If the optimal value of the variable \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\hat{a}}_{k}}\left( n \right)$$\end{document} is not binary, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\lfloor \hat{a}_k\left( n\right) \rceil$$\end{document} can be utilized instead of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\hat{a}}_{k}}\left( n \right)$$\end{document} as the final binary optimized value, where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\lfloor x\rceil$$\end{document} denotes the integer nearest to x. Owing to the limitation imposed by the optimization objective, UAV only maintains the longest transmission duration with the nearest node to reduce system energy consumption, so the discrepancy between \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\lfloor \hat{a}_k\left( n\right) \rceil$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\hat{a}}_{k}}\left( n \right)$$\end{document} is practically insignificant. Furthermore, the slack variable \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$H_{k}^{{}}\left( n \right)$$\end{document} in Eq. (35) is introduced to replace the non-convex part of the left side of C4.35\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} H_{_k}^{2}\left( n\right) \le \tau _{_k}\left( n\right) R_{_k}\left( n\right) \end{aligned}$$\end{document}Then C4 is reformulated as:36\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} C9:\sum \nolimits _{n=1}^{N}{H_{k}^{2}\left( n \right) \ge {{Q}_{K}},\ \ \ }k\in \kappa \end{aligned}$$\end{document}To guarantee that the slacked problem and the original one are equivalent, Eq. (35) must be incorporated as an additional constraint. The problem \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S_{2}P_{3}$$\end{document} is transformed into:37\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \begin{aligned}&{{S}_{2}}{{P}_{4}}:\underset{\left\{ \boldsymbol{A,T,Z} \right\} }{ \text {min}}\,\left\{ {{E}_{sum}} \right\} \\&\quad \quad \ s.t.\ \ C7-C9,(28),(34) \\ \end{aligned} \end{aligned}$$\end{document}Since the constraints Eq. (29) and C9 are still non-convex constraints on the optimization variables, we employ the SCA algorithm to iteratively refine approximations toward the optimal solution of the primal problem until the algorithm converges. Specifically, for constraint Eq. (29), given that the local point of the lth iteration is \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{z}^{\left( l \right) }}\left( n \right)$$\end{document}, a global lower bound on the right-hand side is obtained by calculating the first-order Taylor expansion as follows:38\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} z^{2}\left( n\right) +\frac{\Delta ^{2}\left( n\right) }{v_{0}^{2}}\ge z^{(l)2}\left( n\right) +2z^{(l)}\left( n\right) \left[ z\left( n\right) -z^{(l)}\left( n\right) \right] +\frac{\Delta ^{2}\left( n\right) }{v_{0}^{2}} \end{aligned}$$\end{document}Then constraint Eq. (29) is rewritten as:39\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \frac{T^{4}(n)}{z^{2}(n)}\le z^{(l)2}(n)+2z^{(l)}(n)\biggl [z(n)-z^{(l)}(n)\biggr ]+\frac{\Delta ^{2}(n)}{v_{0}^{2}},\,\forall n \end{aligned}$$\end{document}Similarly, for constraint C9, given that the local point of the lth iteration is \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$H_{k}^{\left( l \right) }\left( n \right)$$\end{document}, a global lower bound on the left-hand can be written as:40\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} H_{k}^{2}(n)\ge H_{k}^{(l)2}(n)+2H_{k}^{(l)2}(n)\Big (H_{k}(n)-H_{k}^{(l)}(n)\Big ) \end{aligned}$$\end{document}Then constraint C9 is rewritten as:41\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \sum _{n=1}^{N}\biggl [H_{k}^{(l)2}\left( n\right) +2H_{k}^{(l)2}(n)\biggl (H_{k}\left( n\right) -H_{k}^{(l)}\left( n\right) \biggr )\biggr ]\ge Q_{K},\, k\in \kappa \end{aligned}$$\end{document}Through the effective handling of the constraints Eq. (29) and C9, the problem \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S_{2}P_{4}$$\end{document} is transformed into:42\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \begin{aligned}&{{S}_{2}}{{P}_{5}}:\underset{\left\{ \boldsymbol{A,T,Z} \right\} }{ \text {min}}\,\left\{ {{E}_{sum}} \right\} \\&s.t.\; \ \ C7,C8, (34), (38), (40) \\ \end{aligned} \end{aligned}$$\end{document}The above procedure converts the problem \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S_{2}P_{4}$$\end{document} into a set of solvable convex sub-problems \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S_{2}P_{5}$$\end{document}. We keep solving these convex sub-problems and iteratively updating \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{z}^{\left( l \right) }}\left( n \right)$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$H_{k}^{\left( l \right) }\left( n \right)$$\end{document} until the the difference in objective value between the two successive iterations falls below or equals to a preset threshold \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\delta }_{1}}$$\end{document}, thereby yielding the optimized UAV flight duration \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\boldsymbol{{T}}}}$$\end{document} and SNs scheduling strategy \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\boldsymbol{{A}}}}$$\end{document}. 2) Optimization of UAV Trajectory: In stage 2, we optimize UAV trajectory \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\boldsymbol{{Q}}}}$$\end{document} based on the outcomes derived from stage 1. The problem \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S_{2}P_{2}$$\end{document} is reformulated as:43\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \begin{aligned}&{{S}_{2}}{{P}_{6}}:\underset{\left\{ \boldsymbol{Q,Z} \right\} }{{\text {min}}}\,\left\{ \omega \sum \limits _{n=1}^{N}{\left[ {{P}_{0}}\left( T\left( n \right) +\frac{3{{\Delta }^{2}}\left( n \right) }{U_{tip}^{2}T\left( n \right) } \right) +{{P}_{1}}z\left( n \right) +\frac{1}{2}{{d}_{0}}\rho uA\frac{{{\Delta }^{3}}\left( n \right) }{{{T}^{2}}\left( n \right) } \right] } \right\} \\&\quad \quad \quad \ s.t.\ C2,C4,C5, (28) \\ \end{aligned} \end{aligned}$$\end{document}This problem is non-convex since C4 and Eq. (29) are non-convex constraints. Then, the slack variable shown in Eq. (44) is introduced.44\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} I_{k}(n)\le \frac{1}{\left\| \boldsymbol{w_k}-{{\boldsymbol{q}}}(n)\right\| ^{2}+H^{2}} \end{aligned}$$\end{document}Then constraint C4 is rewritten as:45\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} C10:B\sum _{n=1}^{N}a_{k}\left( n\right) T\left( n\right) \log _{2}\left( 1+\gamma _{k}I_{k}\left( n\right) \right) \ge Q_{K},k\in \kappa \end{aligned}$$\end{document}And the problem \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S_{2}P_{6}$$\end{document} is transformed into:46\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \begin{aligned}&{{S}_{2}}{{P}_{7}}:\underset{\left\{ \boldsymbol{I,Q,Z} \right\} }{ \text {min}}\,\left\{ \omega \sum \limits _{n=1}^{N}{\left[ {{P}_{0}}\left( T\left( n \right) +\frac{3{{\Delta }^{2}}\left( n \right) }{U_{tip}^{2}T\left( n \right) } \right) +{{P}_{1}}z\left( n \right) +\frac{1}{2}{{d}_{0}}\rho uA\frac{{{\Delta }^{3}}\left( n \right) }{{{T}^{2}}\left( n \right) } \right] } \right\} \\&\quad \quad \quad \ s.t.\ C2,C5,C10, (28), (43) \\ \end{aligned} \end{aligned}$$\end{document}The constraints Eq. (29) and Eq. (44) in the above problem are still non-convex, and similarly, the SCA algorithm is adopted to address the problem \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S_{2}P_{7}$$\end{document}. For constraint Eq. (29), given that the local point of the lth iteration is \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{z}^{\left( l \right) }}\left( n \right)$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\boldsymbol{q}}^{\left( l \right) }}\left( n \right)$$\end{document}, a global lower bound on the right-hand side is denoted as:47\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} z^{2}\left( n\right) +\frac{\Delta ^{2}\left( n\right) }{v_{0}^{2}}\ge J^{\left( l\right) }\left\{ \boldsymbol{q}\left( n\right) ,z\left( n\right) \right\} \end{aligned}$$\end{document}where,48\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \begin{aligned}&{{J}^{(l)}}\left\{ \boldsymbol{q(n)},z\left( n \right) \right\} \triangleq {{z}^{\left( l \right) 2}}\left( n \right) +2{{z}^{\left( l \right) }}\left( n \right) \left[ z\left( n \right) -{{z}^{\left( l \right) }}\left( n \right) \right] -\frac{||\boldsymbol{q}_{{}}^{(l)}\left( n \right) -\boldsymbol{q}_{{}}^{(l)}\left( n-1 \right) |{{|}^{2}}}{v_{0}^{2}} \\&\quad \quad \quad \quad \quad \quad \quad \quad \ \ +\frac{2}{v_{0}^{2}}{{\left( \boldsymbol{q}_{{}}^{(l)}\left( n \right) -\boldsymbol{q}_{{}}^{(l)}\left( n-1 \right) \right) }^{T}}\left( \boldsymbol{q}\left( n \right) -\boldsymbol{q}\left( n-1 \right) \right) \\ \end{aligned} \end{aligned}$$\end{document}Then constraint Eq. (29) is rewritten as:49\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \frac{T^{4}(n)}{z^{2}(n)}\le J^{(l)}\{\boldsymbol{q(n)},z(n)\},\forall n \end{aligned}$$\end{document}Utilizing the SCA algorithm to perform the same transformation on constraint Eq. (44), it is rewritten as:50\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} I_{k}\left( n\right) \le \frac{H^{2}+2\left\| \boldsymbol{q}^{(l)}\left( n\right) -\boldsymbol{w}_{k}\right\| ^{2}-\left\| \boldsymbol{q}\left( n\right) -\boldsymbol{w}_{k}\right\| ^{2}}{\left( \left\| \boldsymbol{q}^{(l)}\left( n\right) -\boldsymbol{w}_{k}\right\| ^{2}+H^{2}\right) ^{2}} \end{aligned}$$\end{document}Through the effective handling of the constraints Eq. (29) and Eq. (44), the problem \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S_{2}P_{7}$$\end{document} is transformed into:51\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \begin{aligned}&{{S}_{2}}{{P}_{8}}:\underset{\left\{ \boldsymbol{I,Q,Z} \right\} }{ \text {min}}\,\left\{ \omega \sum \limits _{n=1}^{N}{\left[ {{P}_{0}}\left( T\left( n \right) +\frac{3{{\Delta }^{2}}\left( n \right) }{U_{tip}^{2}T\left( n \right) } \right) +{{P}_{1}}z\left( n \right) +\frac{1}{2}{{d}_{0}}\rho uA\frac{{{\Delta }^{3}}\left( n \right) }{{{T}^{2}}\left( n \right) } \right] } \right\} \\&\ s.t.\ C2,C5,C10, (48), (49) \\ \end{aligned} \end{aligned}$$\end{document}The problem \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S_{2}P_{8}$$\end{document} is a convex sub-problem. By continually updating \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{z}^{\left( l \right) }}\left( n \right)$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\boldsymbol{q}}^{\left( l \right) }}\left( n \right)$$\end{document}, and iteratively solving the convex sub-problem until the difference in the objective value between two successive iterations falls below or equals the threshold \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\delta }_{2}}$$\end{document}, the SCA algorithm converges, thereby yielding the optimized UAV trajectory \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\boldsymbol{Q}$$\end{document}. 3) Two-Stage Joint Optimization: Based on the problem-solving method employed in the above two stages, in this section, we further elaborate on the concrete implementation of the TSJO algorithm. Specifically, in the ith iteration, based on the UAV flight duration \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\boldsymbol{Q}}^{\left( i-1 \right) }}$$\end{document} obtained from the previous step, the optimal UAV flight duration \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\boldsymbol{T}_{opt}^{\left( i \right) }$$\end{document} and SNs scheduling \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\boldsymbol{A}_{opt}^{\left( i \right) }$$\end{document} are derived by solving \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S_{2}P_{5}$$\end{document}. Subsequently, based on the above obtained optimization results \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\boldsymbol{A}_{opt}^{\left( i \right) }$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\boldsymbol{T}_{opt}^{\left( i \right) }$$\end{document}, the optimal UAV trajectory \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\boldsymbol{Q}_{opt}^{\left( i \right) }$$\end{document} is updated by solving \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S_{2}P_{8}$$\end{document}. Moreover, the optimized UAV trajectory \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\boldsymbol{Q}_{opt}^{\left( i \right) }$$\end{document} is taken as the initial trajectory for a new iteration. This alternating process is repeated until the difference in objective function value in Eq. (26) between two successive iterations falls below or equals a preset threshold \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\delta$$\end{document}, at which point the algorithm loop is terminated and the final optimized variables \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left\{ \boldsymbol{A}_{opt}^{{}},\boldsymbol{T}_{opt}^{{}},\boldsymbol{Q}_{opt}^{{}} \right\}$$\end{document} are obtained. The execution of the TSJO algorithm is presented in Algorithm 2. Algorithm 2Two-stage joint optimization algorithm. Two-stage joint optimization algorithm. Scheme implementation and complexity analysis To support long-term monitoring missions in remote areas, the energy-efficient optimization scheme proposed in this paper adopts a deployment architecture that combines remote planning with local execution, ensuring theoretical performance while also addressing practical engineering feasibility for real-world deployment. Taking forest monitoring in national nature reserves or key ecological functional zones as an example, a large number of wireless sensing nodes are widely deployed across rugged terrains with limited communication infrastructure to continuously collect multidimensional environmental parameters-such as temperature, humidity, soil moisture, light intensity, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$CO_{2}$$\end{document} concentration, and acoustic/vibration signals. These data are critical for forest health assessment, pest and disease early warning, carbon sequestration quantification, and biodiversity conservation. However, due to constraints on energy supply and communication conditions, traditional data backhaul methods struggle to achieve efficient and sustainable data collection. To address the aforementioned challenges, the proposed scheme completes the entire optimization process offline at a forest-edge computing station or a remote data center prior to the start of each monitoring cycle. First, the system employs the CSNS algorithm to solve the node selection sub-problem \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S_{1}P_{2}$$\end{document}, obtaining a sparse set of sampled nodes that satisfies the compressive sensing (CS) reconstruction accuracy requirement while balancing UAV flight distance and energy consumption across the sensor network. Subsequently, based on the obtained node access set, the TSJO algorithm is invoked to solve the energy-efficient iterative optimization sub-problem \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S_{2}P_{2}$$\end{document}, jointly optimizing the UAV trajectory, flight speeds, and communication scheduling strategies for all selected nodes. Nodes not scheduled in this round remain in sleep mode throughout the current mission cycle to conserve their own energy. The optimized results, including the UAV trajectory, flight speeds, and node activation schedule, are transmitted to the UAV via a LEO satellite link. Upon receiving the instructions, the UAV autonomously follows the precomputed route. When the UAV enters the communication coverage range of a scheduled node, it actively transmits an encrypted wake-up beacon. Upon receiving a valid beacon, the node immediately wakes up from sleep mode and uploads its sensed data. After data transmission is complete, the node automatically returns to a low-power sleep mode, effectively avoiding energy waste due to idle listening. Notably, during each mission cycle, the UAV periodically shuttles back and forth between two predefined base stations. The entire process requires no online replanning or high-frequency feedback interaction, significantly reducing dependence on onboard computational resources and real-time communication reliability, making it highly suitable for typical long-term and periodic data collection tasks. Moreover, to respond to unexpected emergency events within the monitoring area, the proposed architecture can be further extended to support a lightweight dynamic priority response mechanism. Specifically, when a sensing node detects an anomalous event, it can actively transmit a short, high-priority urgent beacon. This beacon consumes minimal energy and contains only the event type and the node ID, which is directly reported to the remote ground control station via a LEO satellite link. Upon receiving this request, the control station forcibly includes the node in the sampling set of the CSNS algorithm in the next mission planning cycle and assigns it the highest access priority. The TSJO algorithm then automatically adjusts the UAV’s flight trajectory and time allocation during the subsequent optimization round to prioritize data collection from this emergency node before handling other routine nodes. Once the updated flight instructions are delivered through the satellite network, the UAV autonomously executes the revised mission without any online intervention. This mechanism effectively enhances the system’s responsiveness to non-instantaneous emergency events without significantly increasing overall system overhead. To further analyze the computational complexity of the proposed energy consumption optimization scheme, we separately examine the computational complexities of the CSNS algorithm and the TSJO algorithm.CSNS algorithm: Since the CSNS algorithm essentially solves the optimal node selection problem using the CVX toolbox, according to existing analysis32, its computational complexity is \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$O\left( {{K}^{3.5}} \right)$$\end{document}.TSJO algorithm: Given that both steps 4 and 5 in TSJO algorithm employ the SCA algorithm to approximate the non-convex problem by solving a series of convex sub-problems, the computational complexity of steps 4 and 5 are \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$O\left( {{L}_{1}}{{\left( KN \right) }^{3.5}} \right)$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$O\left( {{L}_{2}}{{N}^{3.5}} \right)$$\end{document}, where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$L_{1}$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$L_{2}$$\end{document} are the number of convex sub-problems solved iteratively in steps 4 and 5, respectively. Let \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$L_{0}$$\end{document} denote the number of iterations for the TSJO algorithm, then the total computational complexity of Algorithm 2 can be computed as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$O\left( {{L}_{0}}\left( {{L}_{1}}{{\left( KN \right) }^{3.5}}+{{L}_{2}}{{N}^{3.5}} \right) \right)$$\end{document}.Since the computational complexity of the energy consumption optimization scheme primarily depends on the CSNS and TSJO algorithms, the overall computational complexity is \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$O\left( L_0\left( L_1\left( KN\right) ^{3.5}+L_2N^{3.5}\right) +K^{3.5}\right)$$\end{document}. When executed on a high-performance computing server, the runtime of this optimization mechanism is significantly shorter than the duration of each mission cycle, thereby satisfying the offline planning requirements for periodic tasks. CSNS algorithm: Since the CSNS algorithm essentially solves the optimal node selection problem using the CVX toolbox, according to existing analysis32, its computational complexity is \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$O\left( {{K}^{3.5}} \right)$$\end{document}. CSNS algorithm: Since the CSNS algorithm essentially solves the optimal node selection problem using the CVX toolbox, according to existing analysis32, its computational complexity is \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$O\left( {{K}^{3.5}} \right)$$\end{document}. TSJO algorithm: Given that both steps 4 and 5 in TSJO algorithm employ the SCA algorithm to approximate the non-convex problem by solving a series of convex sub-problems, the computational complexity of steps 4 and 5 are \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$O\left( {{L}_{1}}{{\left( KN \right) }^{3.5}} \right)$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$O\left( {{L}_{2}}{{N}^{3.5}} \right)$$\end{document}, where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$L_{1}$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$L_{2}$$\end{document} are the number of convex sub-problems solved iteratively in steps 4 and 5, respectively. Let \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$L_{0}$$\end{document} denote the number of iterations for the TSJO algorithm, then the total computational complexity of Algorithm 2 can be computed as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$O\left( {{L}_{0}}\left( {{L}_{1}}{{\left( KN \right) }^{3.5}}+{{L}_{2}}{{N}^{3.5}} \right) \right)$$\end{document}. TSJO algorithm: Given that both steps 4 and 5 in TSJO algorithm employ the SCA algorithm to approximate the non-convex problem by solving a series of convex sub-problems, the computational complexity of steps 4 and 5 are \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$O\left( {{L}_{1}}{{\left( KN \right) }^{3.5}} \right)$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$O\left( {{L}_{2}}{{N}^{3.5}} \right)$$\end{document}, where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$L_{1}$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$L_{2}$$\end{document} are the number of convex sub-problems solved iteratively in steps 4 and 5, respectively. Let \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$L_{0}$$\end{document} denote the number of iterations for the TSJO algorithm, then the total computational complexity of Algorithm 2 can be computed as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$O\left( {{L}_{0}}\left( {{L}_{1}}{{\left( KN \right) }^{3.5}}+{{L}_{2}}{{N}^{3.5}} \right) \right)$$\end{document}. Additionally, the proposed scheme inherently incorporates a lightweight security mechanism, which manifests in two main aspects: 1) CS technology itself provides a certain degree of physical-layer security gain. The original sensing data \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\boldsymbol{z}$$\end{document} is projected into a low-dimensional observation vector \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\boldsymbol{y}=\boldsymbol{\Phi } \boldsymbol{z}$$\end{document}. If the observation matrix \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\boldsymbol{\Phi }$$\end{document} is pre-distributed by the ground control station via a secure satellite link and kept strictly confidential, an adversary who intercepts \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\boldsymbol{y}$$\end{document} cannot reliably reconstruct the original signal \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\boldsymbol{z}$$\end{document}; 2) The system employs an “on-demand wake-up” mechanism, wherein only legitimate nodes scheduled for data collection respond to the UAV’s encrypted wake-up beacon, effectively preventing unauthorized or spoofed nodes from gaining access. To further enhance system security, UAV trajectory and scheduling commands are encrypted before being transmitted over the LEO satellite link, thereby mitigating risks of location exposure or command injection attacks that could arise from open broadcasting. Additionally, sensed data can be protected prior to upload using lightweight cryptographic methods (e.g., AES-128) or physical-layer key generation based on wireless channel characteristics, providing an extra layer of data confidentiality. It should be noted that these security measures introduce no additional energy overhead to the system. Performance evaluations Assume that \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$K\in [40,100]$$\end{document} SNs are randomly dispersed in a two-dimensional monitoring area of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$400\times 400m^2$$\end{document}. Sensory data of SNs is collected by employing a rotary-wing UAV. The initial and final points of UAV are located at \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\boldsymbol{q}_I=(0,200)$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\boldsymbol{q}_F=(400,200)$$\end{document}, respectively. The UAV flight height is fixed to be \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$H=100m$$\end{document}, the maximum flight speed is \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$V_{max}=30m/s$$\end{document}, and the maximum communication coverage angle is set as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\theta =\pi /3$$\end{document}. Table 2 displays the UAV flight parameters11,33. In addition, the UAV trajectory is discretized into \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$N=500$$\end{document} path segments. The transmission power of the SNs is set to \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$P_k=0.5W$$\end{document}. The bandwidth between UAV and SNs is \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$B=1MHz$$\end{document}. The channel noise power variance is set to \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\sigma ^2=-110dBm$$\end{document}, and the reference channel gain at 1 m is \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$h_0=-60dB$$\end{document}. The attenuation factor of NLoS links is taken as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\beta =0.2$$\end{document}. the energy balance factor is set to \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\lambda =2000$$\end{document}.Table 2UAV flight parameters.ParametersValuesParametersValues\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$P_0$$\end{document}\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$79.8\,\textit{W}$$\end{document}\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$d_0$$\end{document}0.6\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$P_1$$\end{document}\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$88.6\,\textit{W}$$\end{document}u0.05\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$U_{tip}$$\end{document}\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$120\,m/s$$\end{document}A\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$0.503\,m^2$$\end{document}\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$v_0$$\end{document}\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$4.03\,m/s$$\end{document}\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\rho$$\end{document}\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$1.225\, kg/m^3$$\end{document} Table 2UAV flight parameters.ParametersValuesParametersValues\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$P_0$$\end{document}\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$79.8\,\textit{W}$$\end{document}\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$d_0$$\end{document}0.6\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$P_1$$\end{document}\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$88.6\,\textit{W}$$\end{document}u0.05\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$U_{tip}$$\end{document}\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$120\,m/s$$\end{document}A\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$0.503\,m^2$$\end{document}\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$v_0$$\end{document}\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$4.03\,m/s$$\end{document}\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\rho$$\end{document}\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$1.225\, kg/m^3$$\end{document} UAV flight parameters. Performance of CSNS algorithm First, we assess the performance of the CSNS algorithm, which is employed to optimize the node selection strategy. The number of SNs in the WSN is set to \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$K=60$$\end{document}. For the objective function in the CSNS algorithm, to shorten the UAV flight distance in each data collection round, UAV needs to frequently visit the nodes located near the connection line between the initial and final points. This method will turn the area around the lines into an access hotspot, thereby disrupting the energy balance in the WSNs. Thus, the objectives of minimizing UAV flight distance and achieving balanced energy consumption are inherently competitive. To quantitatively characterize the competitive relationship between the two objectives, the objective of network energy consumption balance is defined as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$U=\left( \sum \nolimits _{i=1}^{M}{{{U}_{r}}\left( {{k}_{i}} \right) } \right) /M$$\end{document}, where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{k}_{i}}$$\end{document} indicates the node ID selected to participate in data collection. When the node selection strategy makes the objective function value U smaller, the network energy consumption tends to be more balanced. Regarding the total UAV flight distance, once the set of nodes participating in data collection are determined, the objective of minimizing the total flight distance D can be modeled as a Traveling Salesman Problem (TSP). We employ the classical nearest-neighbor heuristic algorithm34 to solve this TSP in this paper: starting from an initial node, the algorithm iteratively selects the nearest unvisited node as the next destination until all nodes have been visited. This approach is computationally efficient and is widely used in real-time applications such as UAV path planning. Under varying weight factors \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\omega }_{0}}\in \left[ 0,1 \right]$$\end{document}, by continuously implementing the CSNS algorithm to solve problem \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S_{1}P_{2}$$\end{document}, a series of corresponding node selection optimization strategies can be obtained. Based on the strategy set obtained above, the network energy consumption balance objective U and the total UAV flight distance objective D can be further calculated.Fig. 3Pareto front for D and U with different M. Pareto front for D and U with different M. Figure 3 the Pareto front of U and D, where the sampling interval of the weight factor \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\omega }_{0}}$$\end{document} is set to 0.1, encompassing 11 sampling points. It is noted that there is some data overlap at certain sample points within the experimental results. We can observe that under different M values, as the weight value \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\omega }_{0}}$$\end{document} increases, the total UAV flight distance objective D gradually increases, while the energy consumption balance objective U gradually decreases, thereby verifying the competitive relationship between U and D. Therefore, in the case of applications demands sensitive to the UAV flight energy consumption, it is necessary to appropriately decrease the weight factor \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\omega }_{0}}$$\end{document}. Conversely, when there is a large deviation in the energy consumption distribution across the network, the weight factor \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\omega }_{0}}$$\end{document} should be increased in time to prevent the energy hole problem caused by the exhaustion of the energy of some SNs. In the following simulations, the weight factor is set to \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\omega }_{0}}=0.5$$\end{document} by default. To further comparatively analyze the CS reconstruction accuracy and UAV flight distance of the CSNS algorithm, we selects three baseline methods for comparison: the random algorithm, the greedy algorithm, and the GNSS algorithm23. The random algorithm selects nodes uniformly at random from the network. The greedy algorithm chooses nodes by optimizing solely for the shortest UAV flight distance. The GNSS algorithm also adopts a random node selection strategy but incorporates Variable Neighborhood Search (VNS) during the path planning stage to refine the TSP solution. The real data gathered from the sensor nodes in the GreenOrbs system are used as the simulation data for the comparative experiments35. The Discrete Fourier Transform (DFT) basis is selected as the sparsifying basis for the data, which is defined as:52\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} {{\mathbf {\Psi }}_{DFT}}=\frac{1}{\sqrt{K}}\left[ \begin{matrix} 1 & 1 & 1 & \cdots & 1 \\ 1 & \omega & {{\omega }^{2}} & \cdots & {{\omega }^{K-1}} \\ 1 & {{\omega }^{2}} & {{\omega }^{4}} & \cdots & {{\omega }^{2(K-1)}} \\ \vdots & \vdots & \vdots & \ddots & \vdots \\ 1 & {{\omega }^{K-1}} & {{\omega }^{2(K-1)}} & \cdots & {{\omega }^{(K-1)(K-1)}} \\ \end{matrix} \right] \end{aligned}$$\end{document}where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\omega ={{e}^{-j\frac{2\pi }{K}}}$$\end{document} is a \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$K$$\end{document}-th root of unity. The Orthogonal Matching Pursuit (OMP) algorithm is selected as the CS reconstruction algorithm, and the relative error of the reconstruction data is defined as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{\left\| \boldsymbol{Z}-\boldsymbol{\hat{Z}} \right\| }_{2}}}/{{{\left\| \boldsymbol{Z} \right\| }_{2}}}\;$$\end{document}, where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\boldsymbol{\hat{Z}}$$\end{document} is the CS reconstruction data of K SNs. All comparative experimental data are the average value of 1000 independent repeated experiments.Fig. 4Performance of different algorithms: (a) Comparison of CS reconstruction performance. (b) Comparison of UAV flight distances. Performance of different algorithms: (a) Comparison of CS reconstruction performance. (b) Comparison of UAV flight distances. In Fig. 4a, we investigate the CS reconstruction accuracy performance of three node selection strategies under different M. The reconstruction error of each scheme gradually decreases as the number of participating nodes (M value) increases. When M exceeds 6, the reconstruction errors of the three algorithms tend to be stable. Importantly, the CSNS algorithm exhibits the smallest error under different M, while the greedy algorithm has the largest error. This is because the CSNS algorithm incorporates a threshold constraint on the reconstruction error during node selection, enabling it to prioritize representative nodes that contribute most significantly to signal recovery, thereby achieving more effective sampling. In contrast, the greedy algorithm adopts a shortest-UAV-flight-distance-first strategy, which leads to non-uniform compressive sensing (CS) sampling and consequently yields the highest reconstruction error among all methods. Both the GNSS and Random algorithms employ random node selection mechanisms that do not account for reconstruction error, resulting in reconstruction errors that are higher than those of the proposed CSNS algorithm but lower than those of the greedy algorithm. In Fig. 4b, the total UAV flight distances are depicted for the three algorithms. As M value increases, UAV flight distances of all three algorithms increase to different degrees. When M=6, the CSNS algorithm is about 45% lower than the random algorithm in UAV flight distance, but about 28% higher than the greedy algorithm. The flight distance of the GNSS algorithm lies between that of the Random and CSNS algorithms, confirming that its path optimization provides a certain degree of energy savings. However, since GNSS does not jointly optimize node selection and path planning, its trajectory length remains longer than that achieved by the proposed CSNS algorithm. It should be noted that the greedy algorithm targets only the minimization of flight distance; thus, its resulting UAV flight distance serves merely as a theoretical lower bound, which is impractical in real-world applications due to its significantly higher reconstruction error. Performance of TSJO algorithm Next, we assess the efficiency of the TSJO algorithm in addressing the energy consumption iterative optimization sub-problem \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S_{2}P$$\end{document}. The number of selected sparse SNs is set to \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$M=6$$\end{document} and the transmission task value of each selected node is \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{Q}_{K}}=50\text {Mbits}$$\end{document} by default. To evaluate the convergence of the alternating iteration optimization process in the TSJO algorithm, the first-order difference value of the UAV-SNs system energy consumption is given as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{E}_{D}}\left( i \right) =E\left( i \right) -E\left( i\text {+}1 \right)$$\end{document}. A smaller \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{E}_{D}}\left( i \right)$$\end{document} indicates that the change in energy consumption after the ith iteration optimization is smaller, showing that the algorithm is closer to convergence. The convergence of the TSJO algorithm under various weight coefficients \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\omega$$\end{document} is shown in Fig. 5a. According to the experimental results, it is evident that the TSJO algorithm converges rapidly. When the number of iterations exceeds 8, the first-order difference value approaches zero, indicating that the energy consumption remains virtually unchanged. When the number of iterations is less than 3, the larger the energy consumption weight coefficient \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\omega$$\end{document} is, the larger the change of the first-order difference value is. Since a larger weighting coefficient \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\omega$$\end{document} indicates a higher proportion of UAV energy consumption in the system energy consumption. This result reflects that the alternating iteration process of the TSJO algorithm has a greater impact on the UAV energy consumption and a relatively smaller impact on the SNs energy consumption.Fig. 5Performance of TSJO algorithm: (a) Convergence of TSJO algorithm under different weight coefficients. (b) Impact of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\omega$$\end{document} on energy consumption of UAV and SNs. Performance of TSJO algorithm: (a) Convergence of TSJO algorithm under different weight coefficients. (b) Impact of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\omega$$\end{document} on energy consumption of UAV and SNs. To further evaluate the impact of the weight coefficient \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\omega$$\end{document}, Fig. 5b illustrates the comparative variations in the energy usage of UAV and SNs under different \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\omega$$\end{document}, where the left and right vertical coordinate are the SNs and UAV energy usage, respectively. We notice that the UAV energy usage decreases and the SNs energy usage increases as the weight coefficient \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\omega$$\end{document} increases. This is because increasing \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\omega$$\end{document} will gradually shift the focus of the system energy consumption optimization towards reducing the UAV energy usage, resulting in a system scheduling strategy that prioritizes shortening the UAV flight distance. However, this strategy will make the UAV fly farther away from the SNs, leading to increasd transmission distance and reduced communication rate, so the SNs energy usage increases accordingly. In the actual system operation process, if the residual energy of UAV is low while the SNs have sufficient energy, the weight \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\omega$$\end{document} should be appropriately increased to decrease the UAV energy usage. Otherwise, the weight \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\omega$$\end{document} should be appropriately increased. Performance of proposed energy optimization scheme To comprehensively evaluate the energy efficiency of the proposed optimization scheme, we consider three baseline schemes for performance comparison:TSP-Random: This scheme performs data collection using a classical TSP-based path planning method34. The UAV departs from a starting point, sequentially flies to the locations above the selected sensor nodes to hover and collect data, and finally arrives at a designated endpoint. Node selection follows a purely random strategy, without considering node spatial distribution or energy consumption.TSP-CSNS: This scheme also adopts the TSP-based path planning framework but employs our proposed CSNS algorithm during the node selection phase. This baseline enables a direct evaluation of the performance gain achieved by incorporating the CSNS algorithm into the energy optimization scheme.GNSS23: This scheme also selects the SNs to be visited randomly, but during the path optimization phase, it employs the VNS algorithm to solve the TSP, aiming to obtain a flight trajectory superior to that produced by conventional TSP methods.In the simulation experiments, the data collection volume per node is set to \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$Q_{K}=50Mbits$$\end{document}, and the energy consumption weighting coefficient is set to \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\omega =0.5$$\end{document}. Figure 6 describes the energy consumption comparison under different network scales, where the number of selected sparse SNs is set to \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$M=0.1K$$\end{document}. The results indicate that the UAV-SNs energy consumption of these four schemes would rise with an increase in network scale. Our proposed scheme uses substantially less energy than the other three benchmark schemes for the same network size. The TSP-Random scheme employs a random node selection strategy and lacks path optimization, resulting in the longest flight distance and the highest energy consumption. In contrast, the TSP-CSNS scheme leverages the proposed CSNS algorithm to optimize node selection, thereby reducing redundant visits and achieving lower energy consumption than TSP-Random, which validates the critical role of intelligent node selection in energy efficiency. The GNSS scheme, although it applied the VNS algorithm during path optimization to improve TSP solution quality and effectively shorten the flight trajectory, still relies on random node selection and fails to jointly consider node distribution and energy-aware coordination; consequently, its energy consumption lies between that of TSP-CSNS and TSP-Random. These simulation results demonstrate that optimizing either the path alone or the node selection alone can yield certain energy savings, but the joint optimization of node selection and path planning proposed in this work achieves significantly greater energy reduction.Fig. 6Evaluation of UAV-SNs energy consumption for three schemes. TSP-Random: This scheme performs data collection using a classical TSP-based path planning method34. The UAV departs from a starting point, sequentially flies to the locations above the selected sensor nodes to hover and collect data, and finally arrives at a designated endpoint. Node selection follows a purely random strategy, without considering node spatial distribution or energy consumption. TSP-Random: This scheme performs data collection using a classical TSP-based path planning method34. The UAV departs from a starting point, sequentially flies to the locations above the selected sensor nodes to hover and collect data, and finally arrives at a designated endpoint. Node selection follows a purely random strategy, without considering node spatial distribution or energy consumption. TSP-CSNS: This scheme also adopts the TSP-based path planning framework but employs our proposed CSNS algorithm during the node selection phase. This baseline enables a direct evaluation of the performance gain achieved by incorporating the CSNS algorithm into the energy optimization scheme. TSP-CSNS: This scheme also adopts the TSP-based path planning framework but employs our proposed CSNS algorithm during the node selection phase. This baseline enables a direct evaluation of the performance gain achieved by incorporating the CSNS algorithm into the energy optimization scheme. GNSS23: This scheme also selects the SNs to be visited randomly, but during the path optimization phase, it employs the VNS algorithm to solve the TSP, aiming to obtain a flight trajectory superior to that produced by conventional TSP methods. GNSS23: This scheme also selects the SNs to be visited randomly, but during the path optimization phase, it employs the VNS algorithm to solve the TSP, aiming to obtain a flight trajectory superior to that produced by conventional TSP methods. Evaluation of UAV-SNs energy consumption for three schemes. Figures 7 and 8 display the UAV flight durations, UAV trajectories, and the SNs scheduling strategies of our proposed energy optimization scheme at different transmission task values, where the number of SNs and the selected sparse SNs are set to \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$K=60$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$M=0.1K$$\end{document}, respectively. As shown in the experimental results in Fig. 7, although the data volume \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{Q}_{K}}$$\end{document} increases from \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$10Mbits$$\end{document} to \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$200Mbits$$\end{document}, the UAV trajectory remains largely unchanged in terms of overall topology and exhibits little visual difference. This is because the UAV is constrained by its maximum communication radius and must always cover the communication range of each scheduled node. However, as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{Q}_{K}}$$\end{document} increases, the UAV tends to fly closer to the nodes to meet the higher data transmission requirements and reduce communication energy consumption, resulting in a slight increase in total flight distance. When the transmission task value is set as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{Q}_{K}}=200Mbits$$\end{document}, the UAV almost needs to fly nearly above the SNs to gather data, which helps improve the communication rate and ensure the effective execution of the data collection task. Notably, although the changes in the flight trajectory are not significant, the total flight time of the UAV rises significantly from 38.9 seconds to 99.2 seconds. This phenomenon stems from the system’s adaptive adjustment in the time domain. The required data transmission time per node increases substantially with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{Q}_{K}}$$\end{document}. To provide sufficient communication windows, the UAV reduces its flight speed when passing near scheduled nodes, ensuring reliable data upload. This approach offers a notable advantage over the time discretization approach which is commonly employed in most UAV trajectory optimization problems.Fig. 7UAV flight trajectories and UAV flight durations under different transmission task values. UAV flight trajectories and UAV flight durations under different transmission task values. Figure 8 presents the SNs scheduling strategies of the selected sparse SNs under four transmission task values. Note that the time range of the horizontal coordinates in different subfigures is not consistent because the total UAV flight duration is different under different transmission task values. Combined with the UAV trajectories in Fig. 7, it can be found that when the transmission task value is determined, the closer the UAV trajectory is to the SNs location, the shorter the communication time of the SNs. This reduction in communication time is attributed to the enhanced communication rates resulting from shorter communication distances. Additionally, owing to the constraint C4 in problem \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S_{2}P_{1}$$\end{document}, the communication time of the SNs increases as the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{Q}_{K}}$$\end{document} value grows.Fig. 8SNs scheduling strategies under different transmission task values. SNs scheduling strategies under different transmission task values. Conclusion The energy consumption of data collection in UAV-assisted WSNs is examined in this work. To decrease the UAV-SNs energy consumption, an energy minimization problem based on the sparse CS data collection technology is analyzed and established, and an effective energy optimization scheme for UAV-assisted data collection is proposed. The intractable energy consumption minimization problem is decomposed into the node selection sub-problem and the energy consumption iterative optimization sub-problem. Then the CSNS algorithm and the TSJO algorithm are designed to solve these two sub-problems respectively, thereby obtaining the approximate solution for the UAV-SNs energy consumption minimization problem. Simulation results validate the advantages of our proposed scheme in terms of both data reconstruction accuracy and system energy consumption. It should be noted that this work adopts a quasi-static LoS channel assumption to clearly reveal the system performance of the proposed scheme under stable conditions. This assumption is well-suited for most long-term, stable, and periodic monitoring scenarios. To address practical deployment challenges such as time-varying channels, link degradation, and increased UAV energy consumption due to wind resistance under adverse weather conditions, future work will incorporate a dynamic adjustment mechanism based on real-time channel feedback. By online estimating and predicting channel states, the system will adaptively optimize the UAV trajectory, scheduling scheduling policy, and communication parameters to enhance robustness in harsh environments. Additionally, we plan to explore event-triggered data collection strategies with dynamic task scheduling and online trajectory replanning to further improve the system’s flexibility and practicality.