A wireless leaf movement sensor system for early detection of abiotic stresses in Zea mays L. Authors: Xu Zhang, Xiang Li, Ming Li, Yumeng Li, Chunying Wang, Haixia Yu, Shidong He, Tingting Zhai, Ping Liu Published: 2025 ABSTRACT AbstractBackgroundAbiotic stresses are detrimental factors for germination, organ development, and other growth activities in maize, which could reduce yield and quality. The analysis of leaf movement is a simple and efficient method to identify stresses as early as possible. This study developed a wireless leaf movement sensor system (WLMS) using a digital inertial measurement unit (IMU) to measure maize leaf movement in real-time and detect abiotic stresses quickly.ResultsThe IMU was designed as a lightweight sensor structure that the IMU was separated from the MCU (microcontroller unit) and connected via flexible cables. This lightweight sensor attached to maize leaves easily and measured leaf movement in real-time with high resolution (measured error of ± 0.25°). The IMU collected leaf movement data and transmitted the data wirelessly to the data receiving terminal (host computer). Meanwhile, the data receiving terminal performed linear fitting on the daily leaf movement data to extract the movement characteristics of maize leaves. The WLMS detected abiotic stress in maize based on the leaf movement characteristics under different stress conditions. The results indicated that the WLMS could detect whether maize was under stress within one day of being stressed and identify the specific type of stress within the following 5–7 days, providing a lead time of 2 days compared to other non-destructive methods (including RGB imaging, hyperspectral analysis, and chlorophyll meters).ConclusionsThis sensor system enables the rapid and early detection and identification of abiotic stresses in maize as a low-cost tool for plant phenotype measurement and plant movement measurement.Keywords: Leaf movement, Abiotic stress, Stress detection, Leaf tracking, Leaf sensor system, Precision agriculture Background As one of the main food crops in China, maize (Zea mays L.) is an important support for China’s agricultural economy. However, abiotic stress, notably drought and salt stress, poses a significant threat to its production [1–3]. Drought stress seriously damages the metabolic and physiological processes of maize, such as stunting of plant growth, deterioration of ear traits, and other physiological problems, leading to a significant decline in maize yield [4–6]. As a salt-sensitive plant, maize growth is also affected by salt stress. Under salt stress, maize experiences a reduction in net photosynthetic rate, stomatal conductance, and intercellular CO2 concentration, which consequently affects leaf area, plant height, and dry matter mass [7, 8]. Therefore, early detection techniques for maize abiotic stress could facilitate maize phenotype measurement and field management. And it’s of great significance to improve maize yield and achieve precision agriculture. The main methods of abiotic stress detection are divided into environment-based detection methods, physiological-based detection methods, and morphology-based detection methods in maize [9–11]. Environment-based detection methods primarily utilize local precipitation, temperature, soil humidity, salt content, and other data to determine whether maize is under stress [12]. However, they cannot truly reflect the degree of coercion through indirect data. Physiology-based detection methods primarily determine stress by measuring the contents of various nutrients in maize leaves, grains, and other organs, as well as parameters such as stem flow, water potential, and transpiration rate [13, 14]. Physiology-based detection methods offer high monitoring accuracy, but they are complex to operate, lack timeliness, and may cause varying degrees of damage to maize. So, they are not suitable for field detection. Morphology-based detection methods mainly utilize RGB cameras, depth cameras, hyperspectral cameras, and other sensors to acquire two-dimensional or three-dimensional images or spectral images of maize [15, 16]. By employing image processing and other techniques, these methods could identify maize stress and achieve non-contact measurements. However, image-based detection methods suffer from poor timeliness and significant errors. Hyperspectral spectrum-based detection has high monitoring accuracy but the huge amount of calculation and high cost are the main problems that restrict its popularization. Plant development and cellular processes—including seed germination, leaf movement, stomatal opening and closure, and stress responses—are regulated by the endogenous circadian clock [17, 18]. This internal timekeeping mechanism enables plants to adapt to periodic variations in light, darkness, and temperature resulting from the Earth’s rotation [19, 20]. By synchronizing the phase (timing of activity) and period (duration of a cycle) of internal rhythms with the external day–night cycle, the clock ensures that key biological activities occur at appropriate times relative to environmental rhythms, thereby conferring an adaptive advantage [21, 22]. As one of the outputs regulated by the circadian system, leaf movement represents a visible response to light and other mechanical or environmental stimuli [23]. Such movements can be influenced by alterations in external factors such as water and nutrient availability. For instance, Arabidopsis (Arabidopsis thaliana) enhances ethylene production under low oxygen or low light conditions, which modulates differential cell expansion in the abaxial and adaxial regions of the petiole, leading to upward leaf hyponasty [24]. Similarly, common bean (Phaseolus vulgaris) exhibits hyponastic leaf movement during drought stress, reducing light interception to minimize photodamage and wilting [25]. Given its responsiveness to both endogenous and environmental signals, leaf movement serves as a valuable physical indicator of plant physiological status and biological traits [26, 27]. In crop species, plant movements are broadly categorized into two types: tropic and nastic movements [28]. Tropic movements are directional growth responses to unilateral external stimuli, whereas nastic movements occur in response to changes in the intensity of environmental factors, independent of stimulus direction. One type of nastic movement closely associated with plant development and growth is rhythmic (or sleep) movement, which exhibits regular diel or circadian periodicity [29]. It is also important to note that abiotic stresses can alter circadian rhythms and thereby affect crop growth patterns. Federico et al. and Shogo et al. employed RGB time-series images to estimate leaf area, using changes in leaf area to investigate leaf movement characteristics under different stress conditions in Arabidopsis thaliana and lettuce, respectively [30, 31]. In addition, laser scanning or Light Detection and Ranging (LiDAR) technology has been utilized to measure geometry and movement in several plant species, such as wheat (Triticum aestivum L.) [32], Calathea roseopicta [33], and grapevine (Vitis vinifera L.) canopy [34]. While extensive studies have been conducted on various physiological traits of maize under drought and salt stress, research focusing on the circadian rhythmic movement of maize leaves remains limited. Moreover, despite significant progress in the use of imaging technologies for quantifying leaf movement, challenges such as temporal resolution in data acquisition, algorithmic complexity, as well as equipment and labor costs, continue to pose constraints for achieving detailed and high-throughput measurements of plant organ dynamics. This study designed a wireless leaf movement sensor system (WLMS) to measure maize leaf movement and detect abiotic stress affecting maize. The low-cost sensor system could measure the pitch angle of maize leaves quickly and accurately and draw the leaf movement curve. In addition, the system employs a sensor structure where the measuring module is separated from the main board to reduce the system error caused by the weight of the sensor. Meanwhile, linear fitting was used to obtain the slope, R-squared (R2), and residual sum of squares from the daily leaf movement curves. This method extracted the movement characteristics of maize leaves and enabled the detection of abiotic stress based on those characteristics under different stress conditions. Furthermore, compared with other non-invasive detection methods, this method demonstrated the rapid response of leaf movement and validated its accuracy by detecting physiological indicators. Materials and methods Plant material and experimental design Maize inbred genotypes B73 were used in the study. A total of 55 plants were grown in the following five treatments: soil relative water content was 70%~75% and irrigated with clean water (CK); soil relative water content was 50%~55% and irrigated with clean water (MD); soil relative water content is less than 25% and no more watering (SD); soil relative water content was 70%–75% and irrigated with 50mmol/L NaCl solution (MS); soil relative water content was 70%–75% and irrigated with 100mmol/L NaCl solution (SS) (Table 1). All the plants were initially grown with sufficient water and the drought treatments were established by withholding the water when the maize was entering the jointing stage. Each treatment group consisted of 11 maize plants, with 5 plants used for measuring leaf movement, 3 plants for non-destructive testing in control experiments, and the remaining 3 plants for measuring physiological parameters. Sensors were placed at the same position (10 cm from the petiole) on the same leaf (the fifth leaf) of each maize plant to measure leaf movement. The maize was grown in the growth chamber in the National Key Laboratory of Wheat Improvement (36°11ʹ39.588ʹʹN, 117°7′10.848ʹʹE). The room temperature of the growth chamber was maintained at 23 ℃, the air humidity was maintained at 40% and the photoperiod was 16 h. Light intensity in the chamber was 30klux. The plants were placed in 5.7-liter pots with the PINDSTRUP peat soil. This soil source had a pH range of 5.5–6.5, mixed with a certain proportion of vermiculite and perlite. Table 1Research of abiotic stress types, degree, soil relative water content, concentration of NaCl solutionTreatmentStress degreeSoil relative water content (% of field capacity)The concentration of NaCl substances in irrigation water (mmol/L)CKNo stress70∼750MDMild drought50∼550SDSevere drought≤250MSMild salt stress70∼7550SSSevere salt stress70∼75100 Research of abiotic stress types, degree, soil relative water content, concentration of NaCl solution A total of 60 potted maize plants were cultivated during the trial, each assigned a unique identifier and randomly allocated to one of five treatment groups. Initially, all pots were positioned in a completely randomized arrangement within the growth chamber. Throughout the cultivation period, a systematic chessboard-style rotation was performed every two days to ensure that each plant was exposed to various locations in the chamber, thereby averaging out potential environmental gradients. To meet the stability requirements for WLMS measurements, plants designated for leaf movement measurement were fixed in their final positions at the onset of the stress treatment (21 days after emergence, corresponding to the five-leaf stage with fully expanded fifth leaf) and remained stationary thereafter, while the remaining plants continued to be rotated. Before the experiment, the soil-relative water content of all maize was 75%. Soil relative water content was calculated as the current soil water content expressed as a percentage of the soil water content at field capacity. The current soil water content was measured using a soil moisture meter (Model SN-*-SC-35DC, Environmental Monitoring) with a three-point measurement method. The field capacity was determined by thoroughly saturating the soil and allowing it to drain freely for 48 h, after which it was measured using the same soil moisture meter. The daily irrigation requirement for each treatment group was calculated based on the gravimetric soil water content. Prior to planting, the mass of each empty pot, Gp (g), and the total mass after filling with soil, G0 (g), were measured using an electronic balance. The initial soil water content, M0 (%), was obtained using the soil moisture meter. The dry mass of the soil in each pot, G1 (g), was then derived as Eqs. (1), The daily irrigation amount for each maize plant, D (g), was calculated using Eqs. (2), Where Mr is the target soil water content (%), Ma is the measured actual soil water content (%). Wireless leaf movement sensor system System structure design The WLMS consisted of a data acquisition terminal, data receiving terminal, and inertial measurement unit (IMU) (Fig. 1a). The data acquisition terminal collected the leaf movement data and sent the data to the data receiving terminal. The data receiving terminal could connect one to four data acquisition terminals and obtain data from each receiving terminal through polling and transmitting the data to the computer for storage. Both the data acquisition terminal and data receiving terminal adopted an ESP32-PCIO-D4 MCU to process data and used low-power Bluetooth for communication and data transmission (Fig. 1a). The IMU used in the system was MPU6050 sensors. The MPU6050 is a Micro Electro Mechanical System (MEMS) consisting of a three-axis gyroscope and three-axis accelerometer, which could measure the angular rate and linear acceleration of the object under test. When the sensor attitude changed, the internal spring structure changed causing a change in sensor capacitance, which was converted into an electrical signal. the angle of the measured object could be calculated by reading the electrical signal values from the sensor (Fig. 1a, b). Physical schematic diagram of the WLMS and Kalman filtering effect diagram. a Physical diagram of the WLMS. b Actual working diagram of the WLMS, with (1) IMU, (2) data acquisition terminal, and (3) data receiving terminal. c The maize leaf movement is produced by rotation around the YXZ axis. The pitch was defined as around the leaf-stem boundary axis (up or down)(), roll as folding around the main vein (), and deflection as rotation around the main stem axis (top view)(). d, e Comparison schematic between adding and not adding Kalman filtering. The sensor was placed on a flat surface and measured for 20 h. d Effect diagram without a Kalman filter. e Effect diagram with Kalman filter The WLMS was capable of measuring orientation in two primary directions—pitch and yaw —and one secondary direction, roll (Fig. 1c). The sensor’s coordinate system was defined with the X-axis aligned along the chip’s length (roll), the Y-axis along its width (pitch), and the Z-axis perpendicular to the chip surface (yaw), following the right-hand rule. However, maize leaves were alternate and had no petiole. The circadian rhythmic movement was not obvious in leaves around the main axis direction (yaw ). This was because the leaf movement was mainly affected by the light direction and other environmental conditions in this direction, while the light and other environmental parameters in the growth chamber were constant. Additionally, the movement in the rolling direction (roll ) involved transverse rolling to the midrib in maize leaves [35]. Therefore, the leaf movement in the vertical direction (pitch ) was mainly used to explore the stress responses of maize. Specifically, the IMU sensor was affixed to the abaxial side of the fifth maize leaf, approximately 10 cm from the base, using a breathable fabric-based double-sided tape. The IMU sensor was oriented such that its x-axis was aligned parallel to the leaf vein, ensuring that changes in the measured pitch angle directly reflected the leaf’s elevation or drooping movement. The data acquisition terminal was mounted either directly on the main stem of the maize plant or on a vertical wooden stake positioned adjacent to the plant. The IMU sensor remained on the maize leaf throughout the experiment. All sensors enabled fast and stable measurements and provided continuous data at a time resolution of 1.5 s. After placing the sensor for 14 consecutive days, no significant difference was found between the part where the sensor was placed and the rest of the leaf. After removing the sensors, an inspection of the leaf lamina was conducted, resulting in no obvious damage to the leaves. Data measurement and optimization The MPU-6050 sensor was connected to the ESP32 through the Inter-Integrated Circuit (I2C) interface. The MPU-6050 integrated a 3-axis MEMS gyroscope and a 3-axis MEMS accelerometer, utilizing three 16-bit ADCs for the gyroscope and accelerometer to convert its measured analog signal into an output digital signal. The data acquisition terminal read the accelerometer values (, and ) of the MPU6050 in the X, Y and Z directions. The pitch angle () of the leaf could be calculated as Eq. (3). The MPU-6050 accelerometer could directly calculate the attitude angle of maize leaves. However, errors and drifts existed in the obtained data, and these phenomena became more serious as the measurement time increased (Fig. 1d). To reduce systematic errors, a Kalman filter was added to the data receiving terminal. The Kalman filter is a software filtering method that could make educated predictions about where the system would go next, even in the presence of various disturbances. The Kalman filter estimated angle at the time t is expressed as Eqs. (4), where is the Angle value at the last moment of , is the measured value of the sensor at the moment of t , and is Kalman gain and is expressed as Eqs. (5), where is the average error of MPU6050, and is the covariance matrix of the state matrix at time t estimated from time . are determined by at time , and is the covariance matrix of state matrix , where is the average error of the gyroscope, is the sampling period of the sensor and Q is the process noise covariance of the gyroscope and accelerometer. The sensor was fixed on the plane for 20 h to calculate the average measurement error to verify the effect of the Kalman filter. The results indicated that the sensor utilizing Kalman filtering eliminated data drift and reduced the measurement error to approximately ± 0.25° compared to the sensor that directly read the angle (Fig. 1e). Comparative experiments Physiological trait testing experiment design The study conducted physiological assessments on maize exhibiting distinct drought or salt stress characteristics during stress experiments to validate the accuracy of the WLMS in detecting abiotic stress in maize. The physiological data obtained were then compared with the physiological data from the control group (CK). For the drought stress experiment, leaf water content was determined using the oven-drying method. When severe drought stress (SD) exhibited noticeable stress characteristics such as leaf wilting and curling, samples from mild drought stress (MD), severe drought stress (SD), and the control group (CK) were selected from the same leaf. After weighing the fresh weight (Wf), the samples were immersed in distilled water (deionized water) for 6–8 h. Following removal, excess surface moisture was absorbed using blotting paper, and the saturated fresh weight (Wt) was measured. Subsequently, the samples were blanched at 105 °C for 15 min, and dried at 80 °C for a day to obtain dry weight (Wd), and the relative water content (RWC) was calculated using Eq. (6). Additionally, the study measured the activities of superoxide dismutase (SOD), peroxidase (POD), catalase (CAT), and ascorbate peroxidase (APX) in maize leaves to validate the occurrence of drought stress and salt stress. When severe salt stress (SS) plants exhibited visible stress symptoms (leaf dehydration and wilting), leaf samples (the seventh leaf) were collected from each treatment group according to the following criteria: fully expanded, undamaged, and physiologically healthy. The enzymatic activities of SOD, CAT, APX, and POD were determined by the nitro-blue tetrazolium (NBT) photoreduction method, the hydrogen peroxide decomposition method, the ascorbic acid method, and the guaiacol method, respectively. Non-destructive testing comparative experiments design The non-destructive measurements were conducted daily at 12:00 PM for a duration of 7 days, starting from the initiation of maize stress experiments. RGB images and spectral images were captured using the GaiaField Pro-V10 portable hyperspectral camera manufactured by Double-View Spectral Imaging. RGB images were captured with the RGB lens of the hyperspectral camera under normal growth chamber lighting. Time-lapse images of maize under different stress conditions, captured using an RGB camera, were compared with the phenotypic characteristics observed under those conditions to assess stress responses. Spectral images were captured in the greenhouse using halogen lamp light sources, with a spectral range of 400–1000 nm and a resolution of 3.5 nm. The lens used was CCD and sCMOS. The camera was positioned obliquely 1.5 m from the plants. Calibration images were obtained by capturing whiteboard images and lens-covered images before photo capture. Spectral reflectance images of maize were obtained using a black-and-white correction method after maize image capture. The chlorophyll meter utilized the YLS-501 chlorophyll meter manufactured by Koem for measuring the relative chlorophyll content (SPAD value) in maize leaves. Five random points in the same area were selected in the same maize leaf (the fifth leaf) under various treatment conditions. The SPAD values were recorded using the chlorophyll meter. The average SPAD value was then calculated to determine the relative chlorophyll content for each maize plant under different stress conditions. Hyperspectral images of maize were processed using ENVI 5.3 software for radiometric calibration and subsequent reflectance calculation. A consistent region of approximately 1 cm2 was selected 10 cm from the base of the fifth leaf, with care taken to avoid areas exhibiting significant leaf reorientation during the experiment. The average spectral reflectance () and its standard deviation () within each designated region were calculated and used as the spectral reflectance data for the corresponding plant. Employing the random forest algorithm, a comparative analysis was conducted between and on the first and seventh days of each stress experiment to discern key feature points associated with abiotic stress. Wavelength ranges exhibiting higher feature point density were identified as characteristic bands for abiotic stress in maize. Ultimately, four distinct feature bands were pinpointed: 449.2–455.6 nm, 551–567.7 nm, 579–601 nm, and 746–767 nm. Results Leaf movement monitoring The WLMS consisted of a data acquisition terminal, a data receiving terminal, and an inertial measurement unit (IMU) (Fig. 2a). The data acquisition terminal and IMU were designed as a lightweight structure in which the IMU component was separated from the MCU (microcontroller unit) in data acquisition terminal and connected with a flexible cable to reduce the measurement error caused by the weight of the sensor (Fig. 2a). The weight of the separated IMU was only 0.5 g. The leaf movement data was transmitted from the data acquisition terminal to the data receiving terminal using Bluetooth communication and subsequently sent to the host computer via a serial port. Based on the real-time data recorded by the WLMS, the vertical movement of maize leaves was detected. However, during actual operation, the system struggled to accurately measure the period and amplitude of leaf movement in the raw data due to system errors caused by the sensor itself and occasional random errors from the environment. To solve the problem, the Statistical Outlier Removal method was used to remove random errors caused by the environment. Additionally, the data was down-sampled to reduce the influence of systematic errors. The denoised and filtered data have a more obvious circadian rhythm than the raw data (Fig. 2b). WLMS structure and data preprocessing. a Schematic diagram illustrates the structure of the WLMS, with key components including the Inertial Measurement Unit (IMU), data acquisition terminal and data receiving terminal. The IMU was placed on the back of the maize leaf to measure the leaf’s rhythmic movement. The data acquisition terminal received angle data from the IMU and transmitted it wirelessly to the data receiving terminal. Subsequently, the data receiving terminal transferred the raw data to the host computer. b Using the WLMS to measure the movement of maize leaves under laboratory conditions was affected by environmental factors as well as slight errors inherent to the sensor itself (left). After preprocessing by the host computer, smoother rhythmic movement data could be obtained (right) The maize inbred line B73 was used to set up five experimental groups: a control group (CK), a mild drought stress group (MD), a severe drought stress group (SD), a mild salt stress group (MS), and a severe salt stress group (SS). The WLMS was used to measure the movement of maize leaves (the fifth leaf). The leaf movement period for each treatment group ranged from 23.69 to 24.02 h. The mean periods were 23.69 h and 23.81 h in MD and SD under drought stress, respectively. The average periods were 23.82 h and 24.02 h under salt stress of dealing with the MS and SS. The mean period could not be measured due to the absence of rhythmic movement in CK. The results showed that maize showed no significant difference in periodic movement under different stress conditions with the same genotype and the movement period of CK was not detected due to its movement characteristics (Fig. 3a). The rhythmic movement periods and movement curves of maize leaves under five different stress conditions. a Rhythmic movement periods of maize leaves under different stress conditions were calculated by the time intervals between peaks in movement curves. Box plots show the 25% quantile, median (line), and 75% quantile. The whiskers plots represent minimum and maximum values. b Leaf movement curves under non-stress treatment (CK) with soil relative moisture content maintained at 70% to 75%. c Leaf movement curves under mild drought stress (MD) with soil relative moisture content maintained at 50% to 55%. d Leaf movement curves under severe drought stress (SD) with soil relative moisture content below 25%. e Leaf movement curves under mild salt stress (MS) with soil relative moisture content maintained at 70% to 75% and irrigation with a 50 mmol/L NaCl solution. f Leaf movement curves under severe salt stress (SS) with soil relative moisture content maintained at 70% to 75% and irrigation with a 100 mmol/L NaCl solution For maize at the jointing stage, leaf movement could be decomposed into rhythmic movement and growth movement. The rhythmic movement has obvious rhythm and periodicity, while the latter indicates the overall movement trend of leaves. The growth movement reflects the trend of leaf movement (upward or downward) during the leaf growth process. In fact, rhythmic movements do not dominate in maize leaf movement. This is determined by the morphological and growth characteristics of maize leaves. Because the leaf sheath part of maize is tightly wrapped in the main stem growth and the growth trend is radial in maize leaves. Therefore, the leaf movement of maize presented a gradual downward movement and a slight rhythmic movement compared with the leaf movement of Arabidopsis and tomato. This situation was well reflected in CK (Fig. 3b). During the 8-day observation period under control (CK) conditions, the maize leaves exhibited a gradual decline in movement amplitude, with a total range of 10.35° (averaging approximately 1.15° per day). No distinct rhythmic movement was observed, which may be attributed to an oscillation amplitude below 0.5°—the effective detection limit of the WLMS system. This observation also suggests that maize leaves grown under favorable environmental conditions maintain a relatively compact architecture with limited movement dynamics. Under stress conditions, the movement presented unique characteristics in maize leaves (Fig. 3c–f). The maize showed obvious rhythmic movement, which was opposite to the rhythmic movement of tomato, Arabidopsis and banana leaves (up during the day and down at night). The diurnal movement moved up after losing light and down after restoring light in maize leaves. This may be because maize leaves are “patulous” under normal conditions but “closed” upward under light loss and stress conditions. Maize showed no obvious difference in visual observation during 8 days in MD but showed obvious rhythm in leaf movement. The amplitude gradually increased from 1.84° to 3.44° in rhythmic movement along with the drought time (Fig. 3c). At the same time, mild drought stress affected the leaf growth movement to a certain extent. The trend of leaf uplift at night was slightly greater than the decline of leaf movement during the day, contributing to the overall upward trend of leaf movement (the growth movement range was 5.24°). However, in SD the leaf sheath was unable to support the weight of the leaves as the duration of dry periods increased, attributable to the significant water shortage in the soil. The maize leaves showed an accelerated decline trend. In addition, the leaves began to turn wilted on the 7th day of treatment and the movement showed a precipice decline (the movement range of leaves was about 63.69°) in maize leaves (Fig. 3d). The amplitude gradually increased in rhythmic movement, but it only increased from 0.35° to 2.23° due to severe drought, which was smaller than that observed in MD. In the MS and SS, maize growth and development were significantly inhibited due to its salt sensitivity and low tolerance levels [36]. Visually, the leaves become dehydrated and dry from the end, which is more pronounced in older leaves. The leaf movement of maize was similar to that of non-stress treatment in the first few days of MS (Fig. 3e). It showed a slow downward movement and the rhythmic movement was visible but the amplitude was small. The amplitude only increased from 0.28° to 0.73°. As the duration of the salt stress treatment increased, the growth of leaves gradually halted and they began to shrink due to dehydration. This resulted in a noticeable upward trend in leaf movement. However, the movement range was only 4.64° in MS before leaf contraction. The average circadian rhythm amplitude was only 0.5°. Compared with MS, SS caused more obvious rhythmic movement in maize leaves (Fig. 3f). Different from drought stress, the amplitude of rhythmic movement decreased from 2.81° to 1.32° with the increase in stress time. The growth movement was inhibited, with a range of only 4.64° (excluding the last day). On the 6th day of stress, leaves began to dry out and contract, leading to disordered leaf movement. On the 7th day of stress, leaves started to rise rapidly. Analysis of leaf movement characteristics The differences between different stress treatments were analyzed by comparing the data of leaf circadian rhythm amplitude and growth movement changes (Fig. 4a, b). Among them, the difference between the non-stress treatment and the applied stress was the most significant. Since the maize showed no rhythmic movement under the non-stressed condition, the movement period and the amplitude of rhythmic movement could not be measured. Meanwhile, maize leaves will show a certain circadian rhythm under stress conditions, so the rhythmic movement of maize leaves is a sign to judge whether maize is under stress. Comparison of leaf rhythmic movement under different stress conditions. a Leaf movement curves across stress treatments within a unified coordinate system (with gray background shading indicating the dark period). b The daily leaf inclination angle distribution of maize under each stress treatment condition. c The distribution of R² values for linear fitting of daily rhythmic leaf movements under various stress conditions. d The distribution of the residual sum of squares for linear fitting of daily rhythmic leaf movements under various stress conditions. e The variation in slope for linear fitting of daily rhythmic leaf movements under various stress conditions. b–e all use 9:00 AM as the daily boundary line. b–d Box plots display the 25th percentile, median (line), and 75th percentile, with whiskers representing the maximum and minimum values within 1.5 times the interquartile range, and individual points indicating outliers. f, g The maize abiotic stress detection program, based on the leaf movement sensor system, is applied for both early stress identification (f) and precise identification (g) There were differences in leaf movement under different types of stress. (Fig. 4a, b). Under drought stress, the amplitude of leaf rhythmic movements increases with time. The amplitude of leaf rhythmic movements increased from 1.8° to 3.2° under MD. While under SD, it increased from 0.24° to 3.09°. However, under salt stress, the amplitude of leaf rhythmic movements either decreased or showed no significant change over time. The amplitude of rhythmic movements increased from 0.28° to 0.66° under MS, whereas it decreased from 2.8° to 1.1° under SS. Additionally, in terms of the overall leaf movement trend, leaf movement under MD showed a slow upward trend with a range of only 5.24°, while SD exhibited a trend of accelerated decline with a movement range of 63.64°. Leaf movement under MS showed a slow decline initially (first three days), followed by an upward trend and accelerated increase on the seventh day, whereas SS showed an upward trend from the beginning and accelerated upward movement by the seventh day. Based on the aforementioned characteristics, the study performed linear fitting of daily leaf movement curves (with leaf inclination angle in degrees on the Y-axis and time series sequence numbers ranging from 1 to 537 on the X-axis) and obtained the slope (k), R2, and residual sum of squares (v2) of the fitted curves under each stress condition, as shown in Fig. 3c-e. The results indicate that the degree of fitting is high between the trajectory of leaf movement without rhythmic movements and the fitted curve when maize is not under stress conditions, with R2 values all greater than 0.75 and v2 values all less than 5. However, when maize is under stress conditions, the degree of fitting is low between the trajectory of leaf movement with rhythmic movements and the fitted curve (most R2 values are below 0.75 and all v2 values are greater than 5). Therefore, by performing linear fitting on the daily leaf movement trajectory and determining whether v2 is less than 5 and R2 is greater than 0.75 (v2CK < 5 and R2CK >0.75), it is possible to determine whether maize is under stress. This method enables the identification of maize stress status within a single day. Based on the daily fitted parameters k and v2, different stress types can be distinguished (Fig. 4d, e). Specifically, under MD conditions, kMD remains within a narrow range and is consistently greater than 0 (0 ≤ kMD ≤ 0.0065 and v2 >5). In contrast, SD exhibits a decreasing trend in kSD over time, with v2SD spanning a wide range (kSD ≤ 0 and 8 ≤ v2SD ≤ 23500). Under MS, kMS shifts from negative during the first three days to positive in the subsequent four days, showing a sharp increase on the seventh day (kMS (days 1–3) < 0; kMS (days 4–7) > 0; kMS (day 7) ≥ 0.01; and 6 ≤ v2MS ≤ 30). Under SS conditions, all kSS values are positive and increase markedly by the seventh day. Compared to MD, v2SS covers a broader range, with v2SS (day 7) and kSS (day 7) substantially exceeding those in MD (kSS >0, 14 ≤ v2SS ≤ 1320, v2SS (day 7) > 1000, and kSS (day 7) ≥ 0.012). The proposed approach allows detailed discrimination of various stress types, although it requires 5–7 days of leaf movement data for full classification. Abiotic stress detection software According to the aforementioned method, the study developed software for identifying abiotic stress in maize using MATLAB R2018b (uk.mathworks.com/products/matlab) (Fig. 4f, g). The software processes leaf movement data received via a serial port and assesses the status of maize plants based on movement characteristics. Upon detection of rhythmic leaf movement patterns, the system indicates that the plant is under stress, with diagnostic accuracy enhancing as more data are accumulated over time. To evaluate the performance of the software, the WLMS system was deployed on one plant from each of the five treatment groups for real-time monitoring. The results (Fig. 4f, g) demonstrate that the software could reliably detect early stress after only one day of data collection, and was able to correctly identify both the specific stress type and its severity following seven days of continuous measurement. Non-destructive testing comparative experiments To validate the rapidity of the WLMS in detecting abiotic stresses in maize, the study employed three non-destructive measurement methods (RGB camera, chlorophyll meter, and spectrometer) to detect stresses in each experimental group. The identification results based on RGB images are presented in Fig. 5a. During the seven-day stress period, no significant morphological or leaf color differences were observed between the mild drought (MD) and control (CK) groups. In the severe drought stress (SD) group, slight leaf rolling and mild chlorosis became visible in the images captured on the fourth day, with these symptoms progressively intensifying over time. By the seventh day, severe leaf wilting and complete yellowing of the leaf tips were evident in the images. A similar progression of salt stress symptoms was observed in both the mild (MS) and severe (SS) salt stress groups. Notably, signs of dehydration and necrosis at the leaf tips were detectable from the third day of the experiment. Results of the control experiment for non-destructive detection of abiotic stress in maize. a Delayed images of each experimental group after stress application using RGB camera. b Using a handheld spectrometer, SPAD values at the same position of each maize group were measured at the same time every day to obtain the SPAD curves over time under different stress conditions. c Using a hyperspectral camera to capture time-series hyperspectral images of each group of maize under the same halogen lamp light source. d After image calibration, the same region was selected for each maize plant, and the average reflectance curve of that region was calculated. The random forest algorithm was used to compare the temporal changes in spectral reflectance under different stress conditions and to identify key features associated with abiotic stress. e Four wavelength ranges (449.2–455.6 nm, 551–567.7 nm, 579–601 nm, and 746–767 nm) exhibiting higher feature point density were selected as characteristic bands for abiotic stress in maize The chlorophyll meter data indicated that SPAD values remained largely stable across most treatments, with one prominent exception (Fig. 5b). In the severe drought (SD) group, a marked decrease from 42.24% to 18.14% was recorded, corresponding to a 57.07% loss relative to baseline. The decline entered an accelerated phase starting on day 4, where the daily reduction rate rose from 5.32%/day (during days 4–5) to 17.02%/day (during days 6–7). By comparison, variations in other groups were modest: the control (CK) varied by 3.74%, mild drought (MD) by 6.04%, and mild salt stress (MS) by 7.34%. The severe salt (SS) group, despite a net change of 13.09% (36.31% relative decrease), exhibited a different pattern—a sudden decrease on day 3 followed by partial recovery. Therefore, the continuous and accelerating decline uniquely identified severe drought, while the overlapping and less dynamic patterns of the other four treatments prevented their clear distinction via simple algorithmic approaches. Hyperspectral imaging revealed four characteristic wavelength ranges (denoted as , , and ) associated with stress responses in maize (Fig. 5c). These correspond to the average reflectance of 449.2–455.6 nm, 551.0–567.7 nm, 579.0–601.0 nm, and 746.0–767.0 nm, respectively (Fig. 5d). Temporal changes in the average reflectance of these bands are shown in Fig. 5e. Maize plants in the control group (CK) exhibited high spectral stability across all four characteristic bands. The mean reflectance values—, , and —remained consistently around a constant level with no discernible temporal trend, and their variation amplitudes were notably smaller (0.034, 0.053, 0.068, and 0.092, respectively) than those under stress conditions. The mild drought (MD) group showed a variation pattern similar to CK, particularly in the η band, where the fluctuation amplitude (= 0.085) was comparable. However, greater variability was observed in the α, β, and δ bands, with variation amplitudes reaching 0.09, 0.13, and 0.13, respectively. In contrast, the severe drought (SD) treatment displayed no significant spectral shifts during the first three days, but distinct trends emerged from day 4 onward. During this period, and showed a marked increasing trend, with mean reflectance rising from 0.490 to 0.413 to 0.762 and 0.710, corresponding to variation amplitudes = 0.272 and = 0.297. The band also exhibited a fluctuating yet overall increase from 0.290 to 0.378 (= 0.143), whereas the band displayed a decreasing trend from 0.608 to 0.517 (= 0.091). For the mild salt stress (MS) group, the , and bands, as well as over the first six days, showed no consistent directional trend. However, the fluctuation amplitudes of and were 0.119 and 0.116, respectively, both exceeding those of CK. A notable spike in reflectance was observed on day 7, with an increase of 0.132 compared to day 6. Under severe salt stress (SS), the and bands showed a rapid increasing trend during the first five days, rising by 0.095 and 0.113, respectively, and subsequently stabilizing at these elevated levels. The and bands also exhibited clear directional trends, with increasing (= 0.168) and decreasing (= 0.143) over the experimental period. Based on the temporal dynamics of hyperspectral reflectance, a diagnostic framework was established for early identification of specific abiotic stresses. Severe drought (SD) can be detected from Day 4 onward, characterized by a sustained increase in the band (551.0–567.7 nm) exceeding + 0.05 reflectance units, accompanied by a concurrent rise in the band (579.0–601.0 nm) and a decrease in the η-band (746.0–767.0 nm). Severe salt stress (SS) is identifiable within the first 5 days through a rapid co-increase in both band (449.2–455.6 nm) and band reflectance, with thresholds of + 0.09 and + 0.10, respectively. Mild salt stress (MS) is indicated by a distinct late-stage spike in the δ-band reflectance (increase > + 0.12) around Day 7, combined with elevated fluctuation amplitudes in the and bands throughout the experiment. Mild drought (MD), although exhibiting greater variability in the , , and bands compared to the control group, lacks a distinct directional trend, making it difficult to unambiguously distinguish from other stress types. Compared with other non-destructive detection methods, the WLMS system demonstrated rapid and precise detection capabilities (Table 2), particularly in early stress identification and the detection of mild drought stress. Owing to the minimal leaf movement observed in maize under optimal growth conditions (CK), the WLMS system was able to determine the presence of stress within 1–2 days of its onset by detecting the emergence of significant rhythmic movement. Based on this approach, WLMS successfully identified mild drought stress, which is challenging to detect using other non-destructive methods. Table 2Minimum days required for stress identification by different detection methodsWLMS systemRGB imageChlorophyll meterHyperspectral imageEarly stress1–2 days3–4 days (excluding mild drought stress)unrecognized2–3 daysMild drought stress5 daysunrecognizedunrecognizedunrecognizedSevere drought stress5 days7 days7 days5 daysMild salt stress7 days7 daysunrecognized7 daysSevere salt stress7 days7 daysunrecognized5 days Minimum days required for stress identification by different detection methods Determination of physiological indexes To validate the accuracy of the WLMS in detecting abiotic stress in maize, the study conducted physiological assessments on maize exhibiting distinct drought or salt stress characteristics during stress experiments. On the 11th day of experimentation, maize plants subjected to severe drought stress (SD) and severe salt stress (SS) exhibited distinct stress characteristics. The activities of superoxide dismutase (SOD), peroxidase (POD), catalase (CAT), and ascorbate peroxidase (APX), along with relative water content (RWC), were assessed in fully expanded, undamaged, and healthy leaves (the 7th leaf) from each treatment group, with three maize plants measured per treatment. The results, summarized in Fig. 6, revealed a significant reduction in leaf water content under severe drought stress, accompanied by a certain degree of water loss under mild drought stress. Furthermore, all stress-treated maize plants demonstrated a marked increase in the content of oxidative enzymes compared to CK. The results indicated that the physiological responses of maize to stress became evident six days after the initiation of the experiment, and the observed responses were consistent with the applied stress conditions. Schematic diagrams of physiological index determination results and plant morphological representations under different stress conditions. a–e Respectively represent the measurement results of APX, POD, APX, SOD, and CAT. The center line represents the median, the empty square box represents the mean, the edges of the box represent the mean ± SE (Standard Error) and the whiskers extend to a range of 1.5 times the interquartile range (IQR). f Plant morphologies and leaf morphologies of the fifth and sixth leaves in each experimental group on the 11th day of the experiment Discussion Plant leaf movement is one of the physical indicators that provide information about the physiological and biological characteristics of plants. The leaf movement curves under different stress conditions show different characteristics [37, 38]. Plant stress could be detected based on the characteristics [39, 40]. In addition, these features tend to appear earlier than visual features. Some vision-based leaf movement measurement studies have also been used to detect crops such as Arabidopsis, cigar tobacco (Nicotiana tabacum L.), soybeans and maize [41–45]. However, these approaches typically require complex computational processing and are often susceptible to issues such as occlusion [46]. The WLMS system developed in the study enables real-time monitoring of maize leaf movement by capturing leaf angles with high resolution (accuracy: 0.01°, error: ±0.25°). Furthermore, the system adopts a split architecture that physically separates the measurement module from the microcontroller unit (MCU), reducing the weight attached to the leaf and thereby improving the ecological relevance of the acquired angle data. By mounting the IMU sensor directly on the leaf with its x-axis aligned parallel to the leaf vein, the system provides direct measurements of absolute leaf angle without being affected by issues such as leaf occlusion, which often limits optical methods. However, certain limitations in the current design need to be addressed. Although a breathable double-sided adhesive was used to attach the sensor, slight yellowing of the attachment area was observed beginning on the 13th day, which progressively intensified over time. By days 20–25, noticeable necrosis developed in the adhered region, indicating that long-term monitoring with the current attachment method may cause physiological damage. Future iterations should consider more breathable, wearable-type materials to minimize such adverse effects on plant tissues. Through controlled stress experiments, we demonstrated that maize leaf movement exhibits distinctive patterns under different types and levels of abiotic stress. The movement trajectories recorded by the WLMS system can be used to diagnose stress status. Diurnal leaf movement rhythms were altered immediately after stress induction. Under drought stress, rhythmic movement amplitude increased with stress duration, while distinguishing between mild and severe drought required identifying whether the growth-related movement showed an accelerating decline. In salt stress treatments, rhythmic changes were less pronounced. Differentiation between mild and severe salt stress was primarily based on growth-related movement trends: mild salt stress showed an initial decline followed by recovery within the first three days, whereas severe salt stress exhibited a consistent upward trend, with both groups displaying a rapid upward shift on the final day. Based on these dynamic features, a stress detection and classification method was developed, which successfully identified early stress within 1–2 days and specific stress types within 5–7 days. Compared to other non-destructive techniques—such as RGB imaging, chlorophyll meters, and hyperspectral imaging—the WLMS system showed considerable advantages in early stress detection and in identifying mild drought stress, while performing comparably in detecting other stress types. It should be noted, however, that advanced computational approaches—such as deep learning-based RGB image analysis or retrieval of stress-sensitive biochemical components from hyperspectral data—could potentially narrow WLMS’s lead in early and mild-stress detection, and may yield better performance in other stress classification tasks. In addition, although fluorescence measurements were not included in the study due to experimental constraints, they are presumed to offer strong potential for early stress detection. Nevertheless, the WLMS system retains distinct advantages in terms of device cost, measurement time, labor requirements, and capability for real-time monitoring. As more leaf movement data are accumulated, more sophisticated algorithms can be developed to further shorten detection time and improve accuracy. Monitoring maize stress via leaf movement offers significant value for both crop management and stress-resilient breeding. Digital motion sensors represent a viable alternative to imaging technologies, particularly in cases where plant leaves are not fully visible. However, the study was limited to pitch angle measurements due to fixed indoor lighting conditions and the specific rolling behavior of maize leaves. Moreover, leaf movement is highly susceptible to environmental factors such as wind, rain, and mechanical damage, indicating that field applications would require careful environmental correction. Deploying the system in the field would also necessitate long-range wireless communication protocols (e.g., LoRa or ZigBee) and sustainable power solutions such as solar-powered battery systems to enable long-term monitoring. Addressing these challenges is essential for transitioning the system to field conditions and represents a key focus of our future research. Conclusions Digital motion sensors serve as a suitable alternative to imaging technologies for monitoring both natural and stress-induced plant movements, particularly in cases where the plant or specific plant organs are not fully visible. Under such conditions, direct sensor contact enables the capture of intricate and dynamic leaf movements. Compared with non-invasive stress detection methods, leaf movement permits faster and more accurate identification of abiotic stress, which holds significant implications for stress detection and management in maize. Furthermore, the lightweight design of the sensor structure effectively minimizes systematic errors caused by sensor weight. Based on these findings, it can be concluded that the WLMS can measure maize leaf movement trajectories in real time and determine stress status based on motion characteristics under different stress conditions. Additionally, this sensor system serves as a low-cost plant phenotyping tool with potential for application in detecting plant movement across other species. Acknowledgements The authors would like to extend their gratitude to their colleagues at Shandong Agricultural University for their assistance with the experiments, as well as to Professor Shuxin Zhang for his valuable help in reviewing the manuscript. Author contributions Xu Zhang: Methodology, Data curation, Formal analysis, Validation, Writing—original draft. Xiang Li: Conceptualization, Supervision, Writing—review & editing. Ming Li: Methodology, Formal analysis. Yumeng Li: Data curation. Chunying Wang: Formal analysis. Haixia Yu: Supervision. Shidong He: Experiment. Tingting Zhai: Formal analysis. Ping Liu: Conceptualization, Supervision, Writing—review & editing, Funding acquisition. Funding This work was supported partially by the NSFC (31871543), Key R&D Program of Shandong Province, China (Nos.2022LZGCQY002, 2021TZXD001 and 2022TZXD0010), the Natural Science Foundation of Shandong (No. ZR2020KF002). Data availability The datasets used and analysed during the current study are available from the corresponding author on reasonable request. Declarations Ethics approval and consent to participate This article does not contain any studies with human or animal subjects. Consent for publication Not applicable. Competing interests The authors declare no competing interests. Footnotes Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Xu Zhang and Xiang Li contributed equally to this work and share first authorship. Contributor Information Xiang Li, Email: lixiang@sdau.edu.cn. Ping Liu, Email: liuping@sdau.edu.cn. References Associated Data Data Availability Statement The datasets used and analysed during the current study are available from the corresponding author on reasonable request.