Advancing Image Super-resolution Techniques in Remote Sensing: A Comprehensive Survey
Yunliang Qi, Meng Lou, Yimin Liu, Lu Li, Zhen Yang, Wen Nie · 2025
Read Advancing Image Super-resolution Techniques in Remote Sensing: A Comprehensive Survey by Yunliang Qi, Meng Lou, Yimin Liu, Lu Li, Zhen Yang, Wen Nie (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2505-23248.txt.
Introduction
Yunliang Qi and five co-authors survey ways to rebuild sharp images from blurred, low-grade ones taken from space or air. Their survey is from two thousand twenty-five.
Such sharp rebuilds are a core task in the use of far-sensed images. Scores of ways have been put forth in past years, but no full, step-by-step review had been made. The team sorts the ways into taught, untaught, and quality-check groups, and sets out the data sets and scores used to judge them. They weigh each group's strengths and flaws. A key gap: fine grain and true shape are lost when the drop in grade is large. They close with paths for new work.
Catalog record
| CREATOR | Yunliang Qi, Meng Lou, Yimin Liu, Lu Li, Zhen Yang, Wen Nie |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 132 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2505-23248
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Advancing Image Super-resolution Techniques in Remote Sensing: A Comprehensive Survey Yunliang Qi, Meng Lou, Yimin Liu, Lu Li, Zhen Yang, Wen Nie Abstract Remote sensing image super-resolution (RSISR) is a crucial task in remote sensing image processing, aiming to reconstruct high-resolution (HR) images from their low-resolution (LR) counterparts. Despite the growing number of RSISR methods proposed in recent years, a systematic and comprehensive review of these methods is still lacking. This paper presents a thorough review of RSISR algorithms, covering methodologies, datasets, and evaluation metrics. We provide an in-depth analysis of RSISR methods, categorizing them into supervised, unsupervised, and quality evaluation approaches, to help researchers understand current trends and challenges. Our review also discusses the strengths, limitations, and inherent challenges of these techniques. Notably, our analysis reveals significant limitations in existing methods, particularly in preserving fine-grained textures and geometric structures under large-scale degradation. Based on these findings, we outline future research directions, highlighting the need for domain-specific architectures and robust evaluation protocols to bridge the gap between synthetic and real-world RSISR scenarios. 1 Introduction Image super-resolution (SR) involves reconstructing high-resolution (HR) images from low-resolution (LR) images [ 1 ] . As a fundamental computer vision task, SR has far-reaching applications in various domains, including remote sensing [ 2 ] , medical imaging [ 3 ] , and security investigation [ 4 ] . However, remote sensing images (RSI) often suffer from limited spatial resolution due to sensor constraints, such as resolution, cost, orbit altitude, and transmission bandwidth. Despite these limitations, high-quality HR images are essential for real-world remote sensing applications. To address this challenge, remote sensing image super-resolution (RSISR) aims to enhance the detailed information of RSI by generating HR images from LR images, as shown in Fig. 1 . RSISR has been extensively applied in land cover mapping [ 5 ] , agricultural monitoring [ 6 ] , disaster assessment [ 7 ] , and urban monitoring [ 8 ] . Typically, RSISR approaches are categorized as single-frame or multi-frame based on input count. Single-frame methods utilize a single LR image to reconstruct a HR output, whereas multi-frame techniques exploit multiple LR observations from diverse perspectives or spectral bands to enhance reconstruction fidelity. Multi-frame SR benefits from complementary information across temporal, multi-view, or multi-sensor data, enabling improved recovery of fine details and high-frequency components. However, practical deployment is constrained by the limited availability of multi-temporal or multi-sensor imagery [ 9 ] and the computational complexity of accurate image registration. Satellite-based applications, for instance, face challenges in acquiring multi-temporal data due to long revisit intervals. Consequently, several studies [ 10 , 11 , 12 , 13 ] have focused on single-frame RSISR to circumvent these limitations. Motivated by these considerations, this survey primarily examines single-frame RSISR methodologies. RSISR has been an active research area for over three decades, with a plethora of algorithms proposed to date. As illustrated in Fig. 2 , the number of publications on RSISR methods has experienced a rapid surge in recent years. Early RSISR approaches can be broadly categorized into three paradigms: interpolation-based methods [ 14 , 15 ] , transform domain-based methods [ 16 , 17 ] , and reconstruction-based methods [ 18 ] . Interpolation-based methods offer the advantages of low computational complexity and real-time implementation, but their performance is often limited by the quality of the LR images, making it challenging to recover high-frequency details and real structures. In contrast, reconstruction-based methods can better preserve the perceptual properties of LR images [ 9 ] , but they often suffer from high computational complexity and dependence on the detail preservation of LR images. Transform domain-based methods, on the other hand, transform the image into another domain (e.g., wavelet domain or sparse representation domain) and perform HR reconstruction by enhancing high-frequency components. Various transform domain-based methods have been proposed, including wavelet domain-based methods [ 16 , 19 ] , sparse representation methods [ 20 , 18 , 21 ] , and hybrid methods [ 22 ] . In recent years, the rapid development of artificial intelligence (AI) has led to the widespread adoption of deep learning techniques in RSISR. A plethora of deep learning-based RSISR methods have emerged, leveraging various models such as convolutional neural networks (CNNs) [ 23 , 24 ] , generative adversarial networks (GANs) [ 25 , 26 , 27 ] , Transformers [ 28 , 29 , 30 ] , and Mamba [ 31 ] to perform RSISR tasks. Figure 1: The benefit of remote sensing images super-resolution. The first row shows the low-resolution images, while the second row shows the results of 4x super-resolution using the SRCNN method [ 32 , 33 ] . As can be seen, the super-resolution images contain richer details and texture information. Despite the numerous papers on RSISR methods published annually, there is a scarcity of comprehensive survey papers that thoroughly summarize these algorithms. While several excellent survey papers on RSISR exist [ 9 , 14 , 34 , 35 , 36 , 2 , 37 , 38 , 25 ] , they remains limitations and challenges. For instance, early review papers primarily focused on conventional RSISR algorithms [ 9 , 14 ] , which are insufficient to meet current needs due to the lack of investigation of deep learning methods. Additionally, some surveys only cover specific types of methods, such as deep learning-based methods [ 35 , 36 , 2 , 38 ] and GAN-based methods [ 25 ] . Liu et al. [ 34 ] conducted an excellent survey on RSISR, but it was published in 2021 and does not cover algorithms published from 2022 to 2025. Figure 2: The number of papers about SR algorithms for RSI since 2014 (according to Web of Science). In this paper, we address these gaps by presenting a detailed, comprehensive, and systematic review of RSISR methods developed over the past two decades, covering more than 300 references. Meanwhile, the background, taxonomy, recent development characteristics, datasets and evaluation metrics in RSISR are also summarized in detail. In summary, the main contributions of this paper lie in the following aspects: 1. A comprehensive review of RSISR methods. We conduct a thorough survey on the development of RSISR algorithms in the past two decades, including methods, datasets, and evaluation metrics. For a clearer discussion, various RSISR algorithms are classified into different categories, and their representative algorithms are introduced in detail. To the best of our knowledge, this is the most comprehensive review of RSISR methods to date. 2. A systematic summary of the development characteristics of RSISR methods. Recent development features are summarized based on an in-depth survey of more than 300 RSISR papers. Further, development trends and challenges can be extrapolated. 3. A detailed discussion of future prospects. Based on a comprehensive investigation and practical experience, future prospects of RSISR methods are discussed. These insights are expected to encourage further exploration of fundamental challenges and inspire broader engagement in this rapidly evolving field. The rest of paper is organized as follows. Section 2 introduces the background and taxonomy of RSISR methods. Section 3 and 4 elaborate on the supervised and unsupervised RSISR methods respectively, followed the recent development characteristics of the RSISR field in Section 5 .