A hyperspectral vegetation water stress diagnosis method, system, medium and device

Abstract

This invention discloses a hyperspectral method, system, medium, and device for diagnosing vegetation water stress, applied in the field of hyperspectral remote sensing technology. The method includes acquiring hyperspectral image data of selected vegetation and actual growth environment data; extracting hyperspectral reflectance information and performing feature analysis on the hyperspectral reflectance information to obtain a high-dimensional spectral feature set; quantifying the correlation between each high-dimensional spectral feature and the actual growth environment data, and constructing a spectral feature training set based on the quantification results; training a pre-constructed vegetation water parameter inversion model to obtain a trained vegetation water parameter inversion model; inputting the hyperspectral reflectance information of the target vegetation into the vegetation water parameter inversion model to obtain growth environment data; and analyzing at least the soil moisture information and leaf water potential information in the growth environment data to obtain the water stress results of the target vegetation. The method of this invention can achieve efficient and accurate diagnosis of vegetation water stress.

Description

Technical Field

[0001] This invention relates to the field of hyperspectral remote sensing technology, and in particular to a hyperspectral method, system, medium, and device for diagnosing vegetation water stress. Background Technology

[0002] The water status of vegetation is a key factor determining its growth, development, ecological function, and landscape effect. When vegetation is subjected to water stress (including drought and waterlogging stress), its physiological metabolic processes are inhibited. Therefore, diagnosing the water stress status of vegetation is of great practical significance for guiding precise irrigation, timely drainage and maintenance, and ensuring healthy vegetation growth.

[0003] In existing technologies, soil moisture meters and plant water potential pressure chambers are used to measure soil moisture in the root zone and leaf water potential, respectively. This is then combined with experienced judgments made by staff based on appearance characteristics such as leaf color and wilting degree to comprehensively assess the water stress status of the vegetation. However, this method has some limitations. On the one hand, measuring soil moisture in the root zone and leaf water potential is cumbersome and inefficient, making it difficult to meet the needs of high-throughput or large-scale monitoring. On the other hand, leaf color judgment is significantly affected by visual factors, and different personnel have individual differences in color identification, making it difficult to establish a unified and objective assessment standard. This leads to large errors in the diagnosis of water stress, hindering timely implementation of effective irrigation or drainage measures and potentially delaying the optimal maintenance period. Summary of the Invention

[0004] This invention provides a hyperspectral method, system, medium, and device for diagnosing vegetation water stress, in order to solve the technical problems of low efficiency and large error in diagnosing vegetation water stress caused by relying on manual experience and physical measurement, so as to achieve efficient and accurate diagnosis of vegetation water stress.

[0005] To address the aforementioned technical problems, this invention provides a hyperspectral method for diagnosing vegetation water stress, comprising: Acquire hyperspectral image data of the selected vegetation and the actual growth environment data of the selected vegetation; Hyperspectral reflectance information is extracted from the hyperspectral image data, and feature analysis is performed on the hyperspectral reflectance information to obtain a high-dimensional spectral feature set; The correlation between each high-dimensional spectral feature in the high-dimensional spectral feature set and the actual growth environment data is quantified, and a spectral feature training set is constructed based on the quantization results; Based on the spectral feature training set and the actual growth environment data, the pre-constructed vegetation water parameter inversion model is trained to obtain the trained vegetation water parameter inversion model. The hyperspectral reflectance information of the target vegetation is input into the vegetation moisture parameter inversion model to obtain the growth environment data of the target vegetation. At least the soil moisture information and leaf water potential information in the growth environment data are analyzed to obtain the water stress results of the target vegetation.

[0006] As one preferred embodiment, the extraction of hyperspectral reflectance information from the hyperspectral image data includes: The hyperspectral image data is subjected to radiometric calibration to obtain the image data to be corrected; Geometric correction processing is performed on the image data to be corrected to obtain standard hyperspectral image data; Pixel radiance data is extracted from the standard hyperspectral image data, and the pixel radiance data is analyzed to obtain the hyperspectral reflectance information.

[0007] As one preferred embodiment, the step of performing feature analysis on the hyperspectral reflectance information to obtain a high-dimensional spectral feature set includes: The hyperspectral reflectance information is denoised to obtain the denoised hyperspectral reflectance information; The denoised hyperspectral reflectance information is subjected to a first transformation process to obtain first-order spectral features; The denoised hyperspectral reflectance information is subjected to a second transformation process to obtain dual-band spectral features; The first-order spectral features and the dual-band spectral features are integrated to obtain the high-dimensional spectral feature set.

[0008] As one preferred embodiment, the quantification of the correlation between each high-dimensional spectral feature in the high-dimensional spectral feature set and the actual growth environment data, and the construction of a spectral feature training set based on the quantization results, includes: Based on the Pearson correlation principle, correlation analysis is performed on each of the high-dimensional spectral features and the actual growth environment data to obtain correlation information. The correlation information is quantified, and the high-dimensional spectral feature set is filtered based on the quantification results to determine the spectral feature training set.

[0009] As one preferred embodiment, the step of training a pre-constructed vegetation moisture parameter inversion model based on the spectral feature training set and the actual growth environment data to obtain the trained vegetation moisture parameter inversion model includes: The spectral feature training set and the actual growth environment data are input into the data decision layer of the vegetation water parameter inversion model for analysis to obtain a nonlinear mapping relationship. Based on the nonlinear mapping relationship, the vegetation moisture parameter inversion model is trained to obtain the trained vegetation moisture parameter inversion model.

[0010] As one preferred embodiment, the step of inputting the hyperspectral reflectance information of the target vegetation into the vegetation moisture parameter inversion model to obtain the growth environment data of the target vegetation includes: Extract the target spectral features from the hyperspectral reflectance information; Based on the decision tree in the vegetation moisture parameter inversion model, the target spectral features are recursively analyzed to obtain the growth environment data of the target vegetation.

[0011] As one preferred embodiment, the step of analyzing at least the soil moisture information and leaf water potential information in the growth environment data to obtain the water stress results of the target vegetation includes: Extract the soil moisture information and leaf water potential information from the growth environment data; Based on the soil moisture threshold, the soil moisture information is analyzed to obtain the first water stress information; Based on the leaf water potential threshold, the leaf water potential information is analyzed to obtain the second water stress information; The first water stress information and the second water stress information are comprehensively evaluated and processed to obtain the water stress result of the target vegetation.

[0012] As one preferred embodiment, acquiring hyperspectral image data of the selected vegetation includes: The hyperspectral video data acquired by the hyperspectral imaging device is analyzed to obtain the first hyperspectral image frame data; Based on the regional information of the selected vegetation, the first hyperspectral image frame data is segmented to obtain the second hyperspectral image frame data. The second hyperspectral image frame data is subjected to environmental filtering to obtain the hyperspectral image data.

[0013] As one preferred embodiment, after obtaining the water stress result, the method further includes: Based on the water stress results, the water stress level of the target vegetation is determined; Based on the water stress level, determine the maintenance information for the target vegetation; The maintenance information is converted into agricultural machinery control commands; The agricultural machinery control commands are executed to maintain the target vegetation.

[0014] To address the aforementioned technical problems, this invention also provides a hyperspectral vegetation water stress diagnostic system, comprising: The data acquisition module is used to acquire hyperspectral image data of the selected vegetation and the actual growth environment data of the selected vegetation; The information extraction module is used to extract hyperspectral reflectance information from the hyperspectral image data and perform feature analysis on the hyperspectral reflectance information to obtain a high-dimensional spectral feature set. The quantification analysis module is used to quantify the correlation between each high-dimensional spectral feature in the high-dimensional spectral feature set and the actual growth environment data, and to construct a spectral feature training set based on the quantification results; The model training module is used to train the pre-constructed vegetation water parameter inversion model based on the spectral feature training set and the actual growth environment data, so as to obtain the trained vegetation water parameter inversion model. The parameter inversion module is used to input the hyperspectral reflectance information of the target vegetation into the vegetation water parameter inversion model to obtain the growth environment data of the target vegetation. The results analysis module is used to analyze at least the soil moisture information and leaf water potential information in the growth environment data to obtain the water stress results of the target vegetation.

[0015] In another aspect, the present invention provides a computer-readable storage medium storing a computer program, wherein when the device containing the computer-readable storage medium executes the computer program, it implements a hyperspectral vegetation water stress diagnosis method as described above.

[0016] In another aspect, the present invention provides a hyperspectral vegetation water stress diagnostic device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor executes the computer program to implement a hyperspectral vegetation water stress diagnostic method as described above.

[0017] Compared to existing technologies, the beneficial effects of this invention are at least one of the following: This invention obtains a high-dimensional spectral feature set by acquiring hyperspectral image data of selected vegetation and extracting hyperspectral reflectance information for feature analysis, thus fully mining and utilizing the rich information of hyperspectral data. This solves the technical problem of existing technologies over-relying on a few preset vegetation indices, leading to the waste of a large number of sensitive spectral features. This invention quantifies the correlation between each high-dimensional spectral feature and actual growth environment data, and constructs a spectral feature training set based on the quantification results, achieving adaptive selection of features most relevant to water stress. This invention trains a vegetation water parameter inversion model based on the spectral feature training set and actual growth environment data, achieving automated, non-contact inversion of vegetation growth environment data, solving the problems of cumbersome operation and low efficiency of existing physical measurement methods. This invention addresses the technical challenges of meeting high-throughput or large-scale monitoring requirements. By inputting the hyperspectral reflectance information of the target vegetation into the inversion model, it obtains growth environment data including soil moisture and leaf water potential information, achieving a synergistic inversion of vegetation physiological state and environmental water conditions. This solves the problem of existing technologies only inverting a single water parameter and lacking sufficient information dimensions. Furthermore, by analyzing soil moisture and leaf water potential information, this invention obtains water stress results, enabling stress level diagnosis based on dual-index quantitative thresholds. This solves the technical problems of existing technologies lacking unified quantitative standards, having highly subjective diagnostic results, and varying from person to person. Finally, this invention achieves unified diagnosis of multiple water stress types, including drought and waterlogging stress, solving the technical problems of existing technologies focusing primarily on single drought stress and lacking systematic diagnostic capabilities for complex stresses such as waterlogging and rapid drought-waterlogging transitions. Attached Figure Description

[0018] Figure 1 This is a flowchart illustrating a hyperspectral method for diagnosing vegetation water stress in one embodiment of the present invention. Figure 2 This is an experimental scenario diagram from one embodiment of the present invention; Figure 3 This is a graph showing the changes in soil moisture for different samples of *Heliotropium indicum* in one embodiment of the present invention; Figure 4 This is a graph showing the soil moisture changes of different Murraya paniculata samples in one embodiment of the present invention; Figure 5 This is a graph showing the relationship between soil moisture and leaf water potential in one embodiment of the present invention; Figure 6 This is a schematic diagram of a hyperspectral vegetation water stress diagnostic system according to one embodiment of the present invention; Figure 7This is a structural block diagram of a hyperspectral vegetation water stress diagnostic device according to one embodiment of the present invention; Figure label: The module includes: 11. Data acquisition module; 12. Information extraction module; 13. Quantitative analysis module; 14. Model training module; 15. Parameter inversion module; 16. Result analysis module; 21. Processor; 22. Memory. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The purpose of providing these embodiments is to make the disclosure of the present invention more thorough and comprehensive. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0020] In the description of this application, the terms "first," "second," "third," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Therefore, a feature defined with "first," "second," "third," etc., may explicitly or implicitly include one or more of that feature. In the description of this application, unless otherwise stated, "a plurality of" means two or more.

[0021] In the description of this application, it should be noted that, unless otherwise expressly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to fixed connections, detachable connections, or integral connections; they can refer to mechanical connections or electrical connections; they can refer to direct connections or indirect connections through an intermediate medium; and they can refer to the internal communication between two components. The terms "vertical," "horizontal," "left," "right," "upper," "lower," and similar expressions used herein are for illustrative purposes only and do not indicate or imply that the system or component referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as limiting the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items. Those skilled in the art can understand the specific meaning of the above terms in this application based on the specific circumstances.

[0022] In the description of this application, it should be noted that, unless otherwise defined, all technical and scientific terms used in this invention have the same meaning as commonly understood by one of ordinary skill in the art. The terminology used in this specification is for the purpose of describing specific embodiments only and is not intended to limit the invention. Those skilled in the art can understand the specific meaning of the above terms in this application based on the specific circumstances.

[0023] Vegetation water status is a core element in regulating plant physiological metabolic processes and maintaining normal growth and development. It also profoundly affects the stability of regional ecosystem structures, the functioning of ecosystem services, and the visual presentation of landscapes. During their growth cycle, vegetation is susceptible to drought stress, waterlogging stress, and complex water stress caused by rapid shifts between drought and waterlogging. Once under water stress, physiological metabolic activities such as photosynthesis, respiration, and nutrient transport are significantly inhibited, leading to symptoms ranging from leaf chlorosis and stunted growth to plant wilting and death. Therefore, accurate and efficient quantitative diagnosis of vegetation water stress can provide a scientific basis for precise irrigation of landscape vegetation and timely drainage and maintenance of waterlogged areas. This has extremely high engineering application and practical guiding value for ensuring healthy vegetation growth, reducing management costs, and maintaining ecological landscape stability.

[0024] In existing technologies, vegetation water stress diagnosis mainly includes two mainstream technical approaches: manual physicochemical measurement and hyperspectral vegetation index analysis. The traditional physicochemical measurement method is the most widely used. The specific procedure is as follows: a portable soil moisture meter is used to set up monitoring points at multiple locations in the root zone of the vegetation, collecting soil profile temperature and humidity data layer by layer to characterize the root soil water supply conditions; simultaneously, using a plant water potential pressure chamber, healthy functional leaves are harvested detached from the plant, and leaf water potential parameters are accurately measured using the pressure balance principle to quantify the degree of physiological water deficiency in the plant; finally, maintenance personnel rely on visual observation of leaf color depth, degree of wilting and curling, and the extent of yellowing spread, combined with measured soil moisture and leaf water potential data, and rely on industry experience to comprehensively assess the vegetation water stress level.

[0025] Meanwhile, existing technologies also commonly employ empirical spectral vegetation index methods for auxiliary diagnosis. Researchers, leveraging prior knowledge in the field, manually select typical water vapor absorption sensitive bands such as 970nm, 1240nm, 1450nm, and 1940nm. Through fixed band ratio calculations and normalized difference mathematical operations, they construct standardized preset vegetation indices such as NDWI, SRWI, WI, and NDVI. Furthermore, they establish statistical regression models between various vegetation indices and water characteristic parameters such as soil moisture, leaf water potential, and leaf water content, indirectly determining the degree of vegetation water stress based on the fitted correlations. This type of method is simple to operate and has a low application threshold, making it a standard technique in the current field of remote sensing monitoring.

[0026] However, all of the aforementioned existing technologies have inherent drawbacks that are difficult to avoid. Traditional physicochemical measurement methods require manual sampling at fixed points and on-site instrument testing, which is cumbersome and time-consuming. The coverage of single-point monitoring is limited, making it unsuitable for the need for simultaneous monitoring of vegetation over a large area and with high throughput. Furthermore, the judgment of leaf appearance depends on subjective human visual perception, which is greatly affected by factors such as personnel's professional experience and lighting conditions. The lack of unified quantitative evaluation standards can easily lead to biased stress diagnosis results and delays in the optimal timing for vegetation irrigation and drainage maintenance. On the other hand, the empirical spectral vegetation index method relies too heavily on manual subjective selection of feature bands and fixed index formulas. It cannot fully explore the massive information dimensions of continuous narrow bands in the hyperspectral spectrum, and it is difficult to adaptively match the sensitive spectral feature combinations of different vegetation species and stress types. The information utilization rate is low and the model's generalization ability is weak, further limiting the accuracy and universality of water stress diagnosis.

[0027] To address the aforementioned technical problems, one embodiment of the present invention provides a hyperspectral method for diagnosing vegetation water stress. For details, please refer to [link to specific documentation]. Figure 1 , Figure 1 The diagram shown is a flowchart of a hyperspectral vegetation water stress diagnosis method according to one embodiment of the present invention, the method including steps S1 to S6: S1. Obtain hyperspectral image data of the selected vegetation and actual growth environment data of the selected vegetation; S2. Extract hyperspectral reflectance information from the hyperspectral image data, and perform feature analysis on the hyperspectral reflectance information to obtain a high-dimensional spectral feature set; S3. Quantify the correlation between each high-dimensional spectral feature in the high-dimensional spectral feature set and the actual growth environment data, and construct a spectral feature training set based on the quantization results; S4. Based on the spectral feature training set and the actual growth environment data, train the pre-constructed vegetation water parameter inversion model to obtain the trained vegetation water parameter inversion model. S5. Input the hyperspectral reflectance information of the target vegetation into the vegetation moisture parameter inversion model to obtain the growth environment data of the target vegetation. S6. At least the soil moisture information and leaf water potential information in the growth environment data are analyzed to obtain the water stress results of the target vegetation.

[0028] In step S1, hyperspectral image data can characterize the internal cell structure, chlorophyll content, leaf water content, and canopy spectral reflectance response characteristics of vegetation leaves from a microscopic spectral dimension, and is a remote sensing feature source for inverting vegetation water status; while actual growth environment data, as ground-measured true value labels, can provide a benchmark reference for subsequent hyperspectral feature screening, model training, parameter inversion, and stress level calibration.

[0029] Preferably, an ASD FieldSpec4 ground cover spectrometer is used as the data acquisition device. The device covers the entire visible, near-infrared, and short-wave infrared band from 350 to 2500 nm. During the data acquisition process, selected standard plants with uniform growth and free from pests and diseases should be chosen as sampling targets. A well-ventilated and transparent rain-proof canopy should be erected to avoid external environmental interference such as rainfall and strong winds, ensuring natural and stable conditions of light, temperature, and humidity. For details, please refer to [link to relevant documentation]. Figure 2 , Figure 2 The diagram illustrates an experimental scenario in one embodiment of the present invention. Data collection was conducted during a stable solar altitude angle period from 11:00 to 14:00 on a clear, cloudless or partly cloudy day. Before collection, instrument radiometric calibration was performed using a standard whiteboard to eliminate the influence of ambient light and system errors. Fully expanded functional leaves from each plant, with different orientations, canopy heights, and uniform growth patterns, were selected as sampling samples. Multiple canopy spectral curves were repeatedly collected from a single sample. Mean smoothing preprocessing was performed on multiple spectral data from the same sample to remove abnormal noise spectral curves. The data was then exported and preprocessed using ViewSpecPro software to obtain hyperspectral image data for subsequent feature extraction and transformation analysis.

[0030] In embodiments of the present invention, the acquisition equipment for vegetation hyperspectral image data is not limited to the ASD FieldSpec4 ground object spectrometer. It can also employ airborne hyperspectral imagers, UAVs equipped with hyperspectral cameras, ground-based fixed hyperspectral video acquisition devices, and other equipment with continuous hyperspectral video acquisition capabilities. Such equipment can continuously acquire hyperspectral video data streams of the target area in real time, enabling long-term, dynamic, and continuous capture of spectral information of the vegetation canopy. Compared to single-point ground object spectrometers, it offers advantages such as large-scale, automated, and continuous temporal monitoring, making it more suitable for large-scale diagnosis of vegetation water stress in regional settings such as parks and woodlands.

[0031] First, the selected vegetation area is continuously filmed using a hyperspectral video acquisition device to obtain raw hyperspectral video data. Then, the hyperspectral video data is analyzed frame-by-frame and decomposed, separating individual raw hyperspectral images from the continuous video stream at fixed time intervals or frame rates. These single raw hyperspectral images obtained after frame-by-frame decomposition constitute the first hyperspectral image frame data. The first hyperspectral image frame data completely preserves the spectral information of the entire scene at the time of video acquisition, including not only the target vegetation canopy but also information on bare soil, shadows, surrounding weeds, background facilities, and other irrelevant features, providing the original basic image source for subsequent accurate segmentation processing.

[0032] Secondly, the selected vegetation specifically refers to typical shrubs and herbaceous plants in South China, preferably native dominant plants such as Murraya paniculata and Clerodendrum trichotomum, which are commonly used in garden ecological restoration and green space creation. These are also the core monitoring targets for water stress diagnosis in this invention. The spatial boundary coordinates and contour range of each target plant or patch are delineated in advance based on on-site aerial photography, on-site contour mapping, and manual vector outlining. Alternatively, image edge detection and contour recognition algorithms are used to automatically extract the edge of the vegetation canopy, generating a mask of the target vegetation area, which serves as the exclusive area information for the vegetation.

[0033] Based on the identified vegetation area information, image segmentation processing is performed on the first hyperspectral image frame data. By using region masking, threshold segmentation, or semantic segmentation, non-target background areas such as soil, shadows, and debris are removed from the image, and only the effective pixel areas where the selected vegetation canopy is located are retained. The vegetation-specific hyperspectral frame image obtained after segmentation and purification is the second hyperspectral image frame data. This frame data has removed irrelevant background interference and focuses on the effective spectral information of the vegetation canopy.

[0034] Furthermore, field data acquisition is susceptible to environmental interference from factors such as light intensity, cloud cover, atmospheric scattering, and light wind disturbance, resulting in image noise, spectral distortion, and uneven brightness. Therefore, environmental filtering processing is required for the second hyperspectral image frame data. Algorithms such as Gaussian filtering, median filtering, and wavelet denoising can be used to smooth random noise, suppress the spectral distortion caused by lighting and atmospheric conditions, and preserve the true spectral reflectance characteristics of the vegetation canopy. The final result is standardized hyperspectral image data with clean image quality, reliable spectral information, and direct applicability for feature extraction and analysis.

[0035] On the other hand, actual growth environment data includes at least soil moisture data and leaf water potential data. During the same period of hyperspectral synchronous acquisition, a TR-6D soil moisture meter was used to randomly select three different monitoring points in the root zone of each plant. Soil temperature and humidity were measured at these points, and the average value of the three measurements was taken as the true soil moisture data for that plant. For extreme conditions of complete soil flooding in the waterlogging stress and drought-flood transition groups, data distorted by the instrument were removed, and only valid soil moisture samples from the natural rain-fed and drought-stress groups were retained for modeling and correlation. Similarly, standard functional leaves of the plant canopy were collected synchronously. Samples were taken from the petiole and branches at flat cuts, immediately sealed and protected from light, and sent to the laboratory. A PMS 1000 portable plant water potential pressure chamber was used to measure the leaf water potential of each leaf according to the pressure balance method, and the average value was taken as the physiological true value data for the corresponding plant. Continuous observations were conducted according to different water stress treatment groups and different time points to form a time-seriesd leaf water potential sample dataset.

[0036] In step S2, hyperspectral image data only contains raw imaging signals such as pixel radiance and grayscale information, which cannot directly characterize the true spectral response characteristics of vegetation itself. It is also susceptible to interference from external factors such as atmospheric scattering, illumination differences, instrument system errors, and surface background. Directly using hyperspectral image data for modeling would introduce a large amount of invalid noise and redundant information, leading to large training biases, low accuracy in water stress retrieval, and poor generalization ability in subsequent models. Therefore, it is necessary to first extract hyperspectral reflectance information that truly reflects the physicochemical properties of vegetation from the hyperspectral image data. Then, through multi-dimensional feature transformation and analytical mining, weak water stress response features hidden within continuous narrow bands are extracted to construct a high-dimensional spectral feature set. This provides a standardized and highly recognizable feature data source for subsequent feature correlation analysis, model training, and quantitative diagnosis of water stress.

[0037] Preferably, the preprocessed hyperspectral image data is radiometrically calibrated and geometrically corrected to convert the image pixel DN values ​​into atmospheric top-layer radiance. Then, combined with the synchronously acquired standard whiteboard radiance data, the whiteboard correction method is used to perform reflectance inversion to eliminate systematic biases caused by illumination angle and atmospheric attenuation. Subsequently, the effective pixels of the selected vegetation canopy are accurately segmented and extracted, and invalid background pixels such as soil, shadows, and weeds are removed. The effective pixels of the canopy are subjected to mean statistics and smoothing and denoising processing to finally obtain hyperspectral reflectance information that can truly represent the spectral response law of the vegetation canopy, covering all continuous bands from visible light and near-infrared to short-wave infrared.

[0038] Furthermore, the denoised and corrected hyperspectral reflectance information is subjected to a first transformation and a second transformation, respectively. The features obtained from the two transformations are then fused and integrated to construct a high-dimensional spectral feature set. The first transformation specifically refers to the first-order derivative transformation of the hyperspectral image, a differential analytical processing performed on the original canopy spectral reflectance curve of the vegetation. The specific implementation process is as follows: based on the continuous narrow band of 350–2500 nm of the hyperspectral imager, the spectral reflectance values ​​and corresponding wavelength values ​​of two adjacent bands are extracted sequentially, and then calculated according to the formula for the first-order derivative of the spectrum. In the formula, X represents the first-order spectral eigenvalue obtained after the first transformation; x i x represents the hyperspectral reflectance value corresponding to the i-th band; i+1 λ represents the hyperspectral reflectance value corresponding to the (i+1)th band adjacent to the ith band; i λ represents the center wavelength corresponding to the i-th band; i+1This represents the center wavelength corresponding to the (i+1)th band. The slope of reflectance variation with wavelength between adjacent bands is calculated. A band-by-band differential operation of the global spectral curve is performed, amplifying and analyzing the reflection peaks, absorption valleys, inflection points, and subtle fluctuations of the spectral curve. This transformation effectively eliminates low-frequency interference such as atmospheric scattering, soil background, and baseline drift, highlighting the subtle deformations and local differences in the spectral curve of vegetation under water stress. After the first transformation, the slope information of reflectance variation between adjacent bands can be obtained, thereby highlighting the reflection peaks, absorption valleys, inflection points, and subtle local fluctuations in the hyperspectral curve, reducing the influence of factors such as light variation, background interference, and baseline drift on the original reflectance information, and improving the accuracy of subsequent extraction of water stress-sensitive features.

[0039] The second transformation process is a dual-band normalization transformation, which is a feature reconstruction method based on normalized difference ratio calculations using pairwise characteristic band combinations. Specifically, the process involves: selecting characteristic band pairs highly correlated with vegetation water physiological responses from the full-band spectrum; choosing the reflectance of any two target bands; and performing standardized calculations using the dual-band normalized difference calculation formula. In the formula, X represents the dual-band spectral characteristic value obtained after the second transformation; x i x represents the hyperspectral reflectance value corresponding to the i-th target band; j This represents the hyperspectral reflectance value corresponding to the j-th target band; where i and j represent the positions of two different bands selected from the hyperspectral reflectance information. Relative spectral difference features between bands are generated; water-sensitive single bands, adjacent narrow bands, or cross-interval feature bands can be freely combined to construct multiple sets of dual-band normalized features in batches. This transformation can weaken the interference of random noise in a single band, strengthen the synergistic response relationship between the two bands, and suppress spectral fluctuations caused by the external environment and instrument measurement errors. After the second transformation, the reflectance difference between the two bands can be normalized, weakening the influence of random noise in a single band, changes in the external environment, and instrument measurement errors, and enhancing the synergistic response features related to vegetation water status between different bands. The resulting dual-band spectral features can be used for subsequent correlation analysis with soil moisture information and leaf water potential information, and further construct a high-dimensional spectral feature set.

[0040] After performing first-order derivative transformation and dual-band normalization transformation respectively, all first-order spectral features and each group of dual-band spectral features are dimensionally concatenated, data normalized, and feature regularized and integrated. Redundant and repetitive features are discarded, and all derived features with strong correlation and high recognizability with vegetation water stress in the two types of transformations are retained. Together, they constitute high-dimensional spectral features with rich dimensions, complete information, and strong anti-interference ability, providing a sufficient and reliable feature input basis for subsequent feature correlation quantitative analysis, model training, and accurate diagnosis of vegetation water stress.

[0041] In step S3, the high-dimensional spectral feature set includes original spectral bands, first-order derivative features, dual-band normalized features, and various spectral vegetation indices. The set contains a large number of features with high dimensional redundancy, and some features have weak correlations with actual growth environment data such as leaf water potential and soil moisture, and are highly interfering. Directly inputting all features into the model for training could easily lead to an excessive number of model parameters, computational redundancy, slow convergence, and overfitting, reducing the model's generalization ability and the accuracy of water stress retrieval. Therefore, it is necessary to quantify the correlation strength between each high-dimensional spectral feature and actual growth environment data, eliminating invalid and weakly correlated redundant features, and retaining strongly correlated sensitive features. This allows for the construction of a concise, effective, and model-appropriate spectral feature training set, laying a high-quality input foundation for subsequent model training.

[0042] In this embodiment of the invention, six spectral vegetation indices with physiological and water characterization significance were selected: Water Vegetation Index (WI), Normalized Difference Water Vegetation Index (NDWI), Water Spectral Index (SRWI), Physiological Reflectance Vegetation Index (PRI), Anthocyanin Reflectance Index (ARI), and Normalized Difference Vegetation Index (NDVI). Table 1 details the calculation formulas for each vegetation index. These indices are incorporated into the overall feature system, forming a complete high-dimensional spectral feature set together with the original single-band hyperspectral features, first-order spectral features, and dual-band spectral features. First, the various features included in the high-dimensional spectral feature set are standardized. All original single-band spectral reflectance features, first-order spectral features obtained through the first transformation, dual-band spectral features obtained through the second transformation, and the six spectral vegetation index features are categorized and arranged according to sample time sequence, vegetation group, and water stress treatment type. These are then uniformly converted into standardized numerical sequences with regular dimensions and consistent units, eliminating differences in numerical magnitude and format between different features and ensuring a unified calculation basis for subsequent statistical analysis.

[0043] Table 1. Formulas for Calculating Six Vegetation Indices Subsequently, data matching and correlation processing was carried out. Each high-dimensional spectral feature numerical sequence was rigorously matched one-to-one with leaf water potential and soil moisture information obtained from synchronous field measurements. This ensured that the spectral feature data corresponding to the same vegetation sample at the same time point and the measured ground growth environment parameters formed a one-to-one pair sample dataset. This guaranteed that the changes in spectral features had a strict temporal and sample-based correspondence with the actual water physiological state of the vegetation and the changes in root soil water environment, providing a real and reliable sample basis for correlation quantitative analysis. For example, Table 2 shows the leaf water potential data of different *Agrostis spp.* samples from March 13 to April 13, and Table 3 shows the leaf water potential data of different *Murraya paniculata* samples from March 13 to April 13. Figure 3 This is a graph showing the changes in soil moisture for different samples of *Gnaphalium affine*. Figure 4 This graph shows the changes in soil moisture for different Murraya paniculata samples. Figure 3 In the text, GN represents auspicious grass nurtured by natural rain, while GD represents auspicious grass subjected to drought stress. Figure 4 In the table, N represents Murraya paniculata under natural rain nourishment, and D represents Murraya paniculata under drought stress. Since the soil of the samples of Murraya paniculata and Murraya paniculata under waterlogging stress and rapid drought-flood transition is in extreme condition, the data measured by the soil moisture measuring instrument at this time is not accurate. Therefore, when analyzing the changes in soil moisture, the soil moisture data of the waterlogging stress group and the rapid drought-flood transition group of Murraya paniculata and Murraya paniculata are removed.

[0044] Table 2. Water potential data of Lucky Grass leaves Table 3 Water potential data for Murraya paniculata leaves Based on this, and using the Pearson correlation principle, feature-by-feature, bivariate statistical analysis was conducted on the paired sample datasets. Each high-dimensional spectral feature was used as the independent variable, and leaf water potential and soil moisture were used as the dependent variables. Pairwise linear correlations were calculated for each feature, yielding the Pearson correlation coefficient R and the significance test P-value. Specifically, please refer to Tables 4 to 6. The correlation coefficient R characterizes the strength and positive / negative correlation trend between the spectral features and actual growth environment parameters, while the significance P-value assesses the statistical reliability and validity of the relationship. Figure 5 The graph shows the relationship between soil moisture and leaf water potential. It can be seen from the graph that the fit between soil moisture and leaf water potential is logarithmic, with R0... 2 It is 0.687.

[0045] Table 4. Correlation between high-dimensional spectral characteristics of *Gynostemma pentaphyllum* and leaf water potential. Table 5. Correlation between high-dimensional spectral characteristics of Murraya paniculata and leaf water potential Table 6. Correlation between high-dimensional spectral characteristics and soil moisture After completing the correlation calculation of all features, the significance P-value is used as the primary screening criterion. Based on the preset significance threshold, it is determined whether the correlation of each feature is significant. Redundant features and invalid interference features that do not meet the significance requirement, have weak intrinsic correlation with leaf water potential and soil moisture, or have no statistical significance are removed. At the same time, combined with the absolute value of the correlation coefficient R, core spectral features with high linear correlation and sensitivity to vegetation water stress response are retained, while redundant features with weak correlation and low contribution are removed.

[0046] Through a complete processing flow including standardization, sample pairing, Pearson correlation quantification, significance testing, and invalid feature removal, high-quality features that are highly correlated with vegetation leaf water potential and soil moisture, statistically significant, and have clear physical meaning are selected and retained from the initial high-dimensional spectral feature set. After integration and regularization, a spectral feature training set with reduced dimensions, high information concentration, strong anti-interference ability, and suitable for model training input standards is formed. This lays a solid data and feature foundation for the accurate training of subsequent vegetation water parameter inversion models, nonlinear relationship fitting, and reliable diagnosis of water stress levels.

[0047] The Pearson correlation principle is a classic statistical analysis method used to measure the degree of linear correlation between two sets of continuous numerical variables. It can quantitatively characterize the strength and positive / negative correlation between high-dimensional spectral characteristic values ​​and actual growth environment data (leaf water potential, soil moisture). It calculates the Pearson correlation coefficient R by dividing the covariance of the variables by their respective standard deviations. The correlation coefficient ranges from -1 to 1; the closer the absolute value of R is to 1, the stronger the linear correlation. A positive R value indicates a positive correlation, and a negative R value indicates a negative correlation. The significance of the correlation can be determined by combining it with the p-value; the smaller the p-value, the higher the reliability of the correlation.

[0048] In step S4, the spectral feature training set after feature filtering contains sensitive information highly correlated with vegetation water stress in the hyperspectral dimension, while the actual growth environment data includes ground-measured ground-based labels such as leaf water potential and soil moisture. The original hyperspectral features and vegetation water parameters do not have a simple linear correspondence; there is a nonlinear correlation caused by complex physiological coupling and environmental disturbances. Linear fitting alone cannot accurately characterize the intrinsic laws governing the changes in spectral features under water stress. Therefore, it is necessary to use the filtered high-quality feature training set combined with ground-based ground-based data to supervise the training of the pre-constructed machine learning model, establishing a stable mapping relationship between high-dimensional spectral features and leaf water potential and soil moisture parameters, enabling the model to automatically invert vegetation water status from unknown hyperspectral data.

[0049] Preferably, the vegetation moisture parameter inversion model pre-constructed in this embodiment of the invention adopts a random forest ensemble learning architecture, which is mainly composed of an input layer, a data decision layer, an ensemble decision tree layer, and an output layer. The input layer is used to receive the normalized spectral feature training set data, complete data normalization, dimension alignment, and sample partitioning, and uniformly deliver standardized feature vectors to subsequent layers. The data decision layer is used as the core correlation analysis layer of the model, responsible for analyzing the inherent correlation between high-dimensional spectral features and the true data of the actual growth environment. The ensemble decision tree layer is composed of multiple independent decision trees in parallel, which perform branching, feature importance scoring, and recursive discrimination on the correlation features output by the data decision layer. The output layer integrates the prediction results of all decision trees, completes regression fitting, and outputs the predicted values ​​of moisture parameters such as vegetation leaf water potential and soil moisture.

[0050] In an embodiment of the present invention, the data decision layer receives spectral feature training set independent variable data pushed by the input layer, as well as actual growth environment data dependent variable true value data composed of leaf water potential and soil moisture. First, it performs one-to-one sample matching, dimensional consistency, and outlier sample removal on the two types of data. Then, it sorts the importance of each spectral feature, performs feature splitting threshold traversal, and further mines feature correlation. By stratifying the feature values, dividing the sample subsets, and continuously analyzing the fluctuation patterns of leaf water potential and soil moisture corresponding to different combinations of spectral features, it mines and constructs a nonlinear mapping relationship from high-dimensional spectral features to water parameters, and characterizes the nonlinear shift of spectral response under water stress, the changes in the depth of absorption valleys, and the complex correlation patterns brought about by canopy structure disturbances.

[0051] Based on the nonlinear mapping relationship obtained from the data decision layer, each decision tree node undergoes binary splitting according to spectral feature thresholds, dividing the sample space layer by layer, fitting the local nonlinear correlation between spectral features and leaf water potential and soil moisture. Multiple decision trees independently complete branch growth and rule construction, forming differentiated decision paths, reducing the risk of overfitting of a single decision tree. The regression prediction results of all decision trees are integrated, and the overall prediction value is output using a weighted average method. The node splitting threshold and feature weights are continuously iteratively optimized. Finally, the coefficient of determination R is used as the final prediction. 2 The root mean square error (RMSE) was used to verify the accuracy. The model parameters were repeatedly fine-tuned until the error between the training set and the test set converged and the fitting effect met the standard. Finally, a vegetation water parameter inversion model that has been trained and can be directly put into engineering application was obtained. Table 7 shows the final model accuracy evaluation results. It can be seen from the table that the vegetation water parameter inversion models constructed from the spectral feature sets of *Ilex cornuta* and *Murraya paniculata* and leaf water potential have high fitting accuracy and low RMSE, indicating that the model fitting effect is good and can be used as a water stress diagnosis model.

[0052] Table 7 Model Accuracy Evaluation In step S5, the vegetation moisture parameter inversion model trained in step S4 has established a stable nonlinear mapping rule between high-dimensional spectral features and actual vegetation moisture physiology and soil environmental parameters. In actual field monitoring scenarios, it is impossible to frequently measure leaf water potential and soil moisture for every target plant using instruments. Manual measurement is time-consuming, labor-intensive, has limited sampling, and cannot monitor large areas simultaneously. Therefore, by inputting the hyperspectral reflectance information of the target vegetation into the fully trained vegetation moisture parameter inversion model, and relying on the spectral-moisture response law learned by the model, the growth environment data corresponding to the target vegetation can be directly inverted from the spectral information. This achieves non-contact, non-destructive, rapid, and automated acquisition of vegetation moisture state parameters, replacing traditional manual physicochemical measurement methods.

[0053] Preferably, the target vegetation hyperspectral reflectance information, collected in real-time in the field and preprocessed with radiometric correction and denoising, is input into the input terminal of the trained vegetation moisture parameter inversion model. The model performs format verification and normalization preprocessing on the received hyperspectral reflectance information, unifies the data units and input dimensions, removes abnormal noise pixels and background interference information, and automatically extracts key target spectral features from the full-band reflectance information according to the screening rules in steps 2 to 3, forming a feature input vector consistent with the model training set structure. The extracted target spectral feature vector is then sent to the model ensemble decision tree layer, which is composed of decision tree nodes. Recursive analysis is conducted using the target spectral features as the root node input. Based on a preset feature splitting threshold, the spectral feature values ​​are classified and divided into intervals, with binary branching recursively splitting layer by layer downwards. Within each node, the response patterns of different spectral features are used to continuously approximate the true numerical ranges of leaf water potential and soil moisture. Through parallel recursive deduction using multiple decision trees, the correlation between local features gradually converges to an overall nonlinear mapping relationship. Finally, the results of all decision tree recursive analysis are integrated, and regression fitting is performed to accurately solve for the leaf water potential and soil moisture parameters of the target vegetation, completing the automated inversion output of growth environment data.

[0054] Among them, growth environment data refers to quantitative parameters that can truly characterize the physiological state of water balance and root water supply conditions of the target vegetation. It is the core basis for subsequent water stress diagnosis and specifically includes soil moisture parameters and leaf water potential parameters.

[0055] In step S6, vegetation water stress includes drought stress, waterlogging stress, and combined stress caused by rapid shifts between drought and waterlogging. Relying solely on soil moisture or leaf water potential cannot comprehensively and objectively determine the true stress state of the vegetation. Soil moisture reflects the water supply environment of the soil where the plant roots are located and is an external environmental indicator; leaf water potential directly reflects the plant's own physiological water surplus / deficit and the degree of stress, and is an internal physiological indicator. These two indicators differ in their representational dimensions and response timelines, and using them alone can easily lead to biased judgments and misjudgments of stress levels. Therefore, it is necessary to extract soil moisture information and leaf water potential information from the growth environment data simultaneously for joint analysis, comprehensively judging from both environmental water supply conditions and plant physiological responses to ultimately obtain objective and reliable results regarding the target vegetation's water stress.

[0056] Preferably, multiple water stress control experiments were conducted, including natural rain-fed, drought, waterlogging, and rapid transition between drought and waterlogging. Soil moisture and corresponding leaf water potential data were continuously collected under different stress gradients. A logarithmic fitting relationship between soil moisture and leaf water potential was constructed. The optimal segmentation method was used to solve for the inflection point of the fitted curve, and the critical soil moisture value corresponding to the critical position of the curve abrupt change was obtained. This value was then verified and calibrated in conjunction with the physiological appearance of the vegetation (leaf curling, yellowing, and inhibited growth), ultimately determining the diagnostic threshold for soil moisture stress, i.e., the soil moisture threshold. The temporal changes in leaf water potential of Murraya paniculata and Clerodendrum trichotomum under different water stress gradients were continuously monitored. The critical leaf water potential values ​​at which the vegetation transitioned from normal growth to mild water stress and severe irreversible stress were statistically analyzed. The ability of the plant to recover its normal physiological state after rehydration was used as a criterion to calibrate the leaf water potential threshold at which the vegetation began to enter water stress and the critical leaf water potential threshold at which it could not recover after stress.

[0057] In an embodiment of the present invention, soil moisture information of the target vegetation root zone is extracted from the growth environment data obtained by inversion, and the measured or model-inverted soil moisture values ​​are compared with preset soil moisture thresholds for judgment; when the soil moisture is lower than the set critical threshold, it is determined that the vegetation root system is not water-supplied enough and there is a risk of drought and water shortage; when the soil moisture is in an excessively saturated state for a long time, it is determined that there is a risk of water accumulation and flooding; according to the degree of deviation of the value from the threshold, it is further divided into different levels of normal, mild drought and severe drought, and a stress judgment conclusion based on the soil moisture dimension is generated, that is, the first water stress information. Extract leaf water potential information from the target plant leaves, and compare the measured or inverted leaf water potential values ​​with preset multi-level leaf water potential critical thresholds one by one. If the leaf water potential is lower than the mild stress threshold, it is determined that the plant has entered a state of water stress at the physiological level. If the leaf water potential further drops below the irreversible severe stress threshold, it is determined that the plant has suffered severe water stress, and even if it is re-watered and drained later, it will be difficult to restore normal growth. The severity of stress is divided according to the deviation from the threshold, forming a state discrimination result based on the plant's physiological level, which is the second water stress information. The first and second water stress information are coupled and cross-validated: soil moisture discrimination results are used as the environmental background basis, and leaf water potential discrimination results are used as the basis for the actual physiological response of vegetation, which are mutually verified and complementary corrections; if soil moisture shows drought and leaf water potential drops below the stress threshold at the same time, it is comprehensively judged as deterministic drought stress; if soil moisture is high and waterlogging occurs and leaf water potential drops at the same time, it is comprehensively judged as waterlogging stress; if there is a time lag difference between the two responses, a compromise correction is made by combining the vegetation growth stage and the stress evolution law; finally, the dual judgment conclusions of environmental indicators and physiological indicators are integrated to uniformly classify the vegetation as normal, mild water stress, moderate water stress, and severe irreversible stress, and output complete and quantitative final water stress results of the target vegetation.

[0058] Preferably, after obtaining the water stress results of the target vegetation, targeted vegetation maintenance information is matched and formulated according to the quantitatively classified water stress level. Then, the semantic maintenance information is converted into electrical control commands that can be recognized by agricultural machinery according to the preset communication protocol and hardware control logic, driving the agricultural machinery to automatically execute the corresponding operation actions, thereby realizing precise, automated, and closed-loop intelligent maintenance management of vegetation.

[0059] This invention categorizes water stress results into multiple levels: normal without stress, mild water stress, moderate water stress, severe irreversible stress, and waterlogging stress. Each level corresponds to a specific maintenance strategy and quantified management parameters. When vegetation is determined to be at the normal without stress level, routine steady-state maintenance information is generated, maintaining the original irrigation cycle and soil moisture range without additional watering or drainage; only basic daily maintenance is required. When vegetation is at the mild drought stress level, micro-irrigation maintenance information is generated, clearly defining the single irrigation water volume, sprinkler coverage area, and staggered operation time. Small-volume, slow irrigation is used to moderately replenish root soil moisture, preventing further exacerbation of stress. When vegetation reaches the moderate drought stress level, quantitative, phased irrigation maintenance information is generated, increasing the single irrigation flow rate, extending the irrigation duration, and setting an interval irrigation plan to gradually restore root zone soil moisture to the suitable growth range for the vegetation, gently repairing the plant's physiological water deficiency. When vegetation is assessed as being under waterlogging stress, specific drainage and maintenance information is generated, specifying the activation of drainage equipment, opening of ditch valves, and setting the duration and flow rate for pumping out accumulated water. This rapidly drains water from the fields, reduces soil saturation, and prevents root necrosis due to waterlogging and oxygen deficiency. When vegetation is under severe irreversible stress, restrictive maintenance information is generated, halting indiscriminate large-scale irrigation and forced drainage operations. The focus shifts to foliar moisture retention, growth monitoring, and marking of weak plants, providing a basis for subsequent vegetation restoration and replanting.

[0060] After determining the vegetation maintenance information, it is further converted into control commands for agricultural machinery. First, the system performs structured parsing of the maintenance information, extracting core quantitative parameters such as operation type, start / stop timing, irrigation flow rate, working duration, and operation area, eliminating redundant descriptive information to form a standardized parameter dataset. Second, following the communication protocols, data frame formats, and electrical control logic of agricultural machinery such as intelligent irrigation equipment, drainage pumps, and garden sprinklers, the parsed maintenance parameters are format-encapsulated and protocol-adapted, converting them into digital signals and control messages that the machinery controller can recognize. Subsequently, corresponding on / off control commands, flow rate adjustment commands, and timed start / stop commands are generated according to the operation type, completing the conversion from text-based maintenance plans to machine-executable control commands. Finally, the generated control commands are sent to the main control unit of the agricultural machinery via wired or wireless IoT communication. After receiving and parsing the commands, the machinery automatically completes the operation processes such as irrigation start / stop, flow rate adjustment, water drainage, and targeted spraying, accurately implementing maintenance measures matching the water stress level, achieving a seamless closed loop between vegetation water stress diagnosis and intelligent maintenance operations.

[0061] Another embodiment of the present invention provides a hyperspectral vegetation water stress diagnostic system. For details, please refer to [link to relevant documentation]. Figure 6 , Figure 6 The diagram shown illustrates the structure of a hyperspectral vegetation water stress diagnostic system according to one embodiment of the present invention. The system includes: Data acquisition module 11 is used to acquire hyperspectral image data of the selected vegetation and actual growth environment data of the selected vegetation; The information extraction module 12 is used to extract hyperspectral reflectance information from the hyperspectral image data and perform feature analysis on the hyperspectral reflectance information to obtain a high-dimensional spectral feature set. The quantification analysis module 13 is used to quantify the correlation between each high-dimensional spectral feature in the high-dimensional spectral feature set and the actual growth environment data, and to construct a spectral feature training set based on the quantification results. The model training module 14 is used to train the pre-constructed vegetation water parameter inversion model based on the spectral feature training set and the actual growth environment data, so as to obtain the trained vegetation water parameter inversion model. The parameter inversion module 15 is used to input the hyperspectral reflectance information of the target vegetation into the vegetation water parameter inversion model to obtain the growth environment data of the target vegetation. The results analysis module 16 is used to analyze at least the soil moisture information and leaf water potential information in the growth environment data to obtain the water stress results of the target vegetation.

[0062] In embodiments of the present invention, the hyperspectral vegetation water stress diagnosis system integrates sensing acquisition, embedded processing, intelligent computing, and result analysis. The information extraction module, as the front-end sensing and processing hardware unit of the system, integrates a hyperspectral imaging chip, an embedded image processor, and a storage cache chip. It undertakes the hardware preprocessing tasks of receiving raw hyperspectral image data, radiometric calibration, atmospheric correction, and geometric correction. At the hardware level, it completes the calculation of hyperspectral reflectance information of the vegetation canopy. At the same time, it has a built-in FPGA high-speed computing chip to realize the denoising of raw spectral data, first derivative transformation, dual-band normalization operation, and solution of multiple vegetation indices. It generates a high-dimensional spectral feature set including single bands, spectral transformation, and vegetation indices from the hardware end, providing a standardized feature data source for back-end analysis.

[0063] Preferably, the data acquisition module is equipped with a ground-based hyperspectral imager, a drone-mounted hyperspectral camera, or a fixed hyperspectral video imager, capable of acquiring real-time hyperspectral image data or continuous hyperspectral video data of the target vegetation area. To eliminate interference factors such as imaging equipment system errors, atmospheric scattering attenuation, and topographic geometric distortion, and to obtain high-precision, high-fidelity vegetation canopy hyperspectral reflectance information, the acquired hyperspectral image data or single-frame image data after video segmentation needs to undergo radiometric calibration, atmospheric correction, and geometric correction preprocessing sequentially to ensure the accuracy of subsequent spectral feature extraction and moisture parameter inversion.

[0064] Radiometric calibration was performed using a combination of standard whiteboard calibration and instrument radiometric calibration coefficients. First, at the field spectral acquisition site, maintaining the same illumination angle and observation height as the vegetation imaging, images were acquired from a standard diffuse reflectance whiteboard to obtain the raw DN grayscale values. Then, the factory-preset radiometric calibration coefficients of the hyperspectral imager were used to establish a linear conversion relationship between the image DN values ​​and the actual entrance pupil radiance. The DN values ​​of each pixel in the raw hyperspectral image were then substituted pixel by pixel into the calibration formula, converting the digitally quantized grayscale signal into physically meaningful atmospheric top-layer radiance. This eliminated inherent system errors such as uneven imaging device response, dark current shift, and lens transmission differences, achieving a standardized conversion of the raw image signal into physical quantities.

[0065] Atmospheric correction was performed on hyperspectral image data using an atmospheric correction algorithm based on a radiative transfer model. Hyperspectral signals acquired in the field are susceptible to atmospheric scattering, water vapor absorption, and aerosol attenuation, leading to distorted spectral reflectance. The specific atmospheric correction process is as follows: Environmental parameters such as atmospheric visibility, solar altitude angle, observation zenith angle, and center latitude and longitude at the time of imaging are input; the radiative transfer model is used to simulate the contributions of atmospheric scattering and absorption; atmospheric path radiation, atmospheric scattered light, and water vapor absorption interference components are stripped band by band from the total radiance of image pixels to restore the true surface reflectance information of the vegetation canopy; the corrected radiance is further converted into the true spectral reflectance of the surface, eliminating spectral distortion caused by the atmospheric environment and restoring the inherent spectral response characteristics of the vegetation itself.

[0066] The purpose of geometric correction is to eliminate image pixel shifts, stretching, distortion, and geographic coordinate misalignment caused by factors such as terrain undulations, shooting tilt angles, UAV attitude drift, and lens distortion during hyperspectral imaging, ensuring accurate matching of image pixels with their actual spatial locations on the ground. The specific process is as follows: First, select clear and stable ground control points in the image, such as road inflections, plot boundaries, and fixed landmarks; accurately extract the pixel coordinates and actual geographic coordinates of each control point in the original distorted image, and construct a geometric correction polynomial fitting model; use the fitting model to resample and correct the position of each pixel in the entire hyperspectral image; use nearest neighbor interpolation or bilinear interpolation algorithms to reconstruct the grayscale and spectral information of the resampled pixels, correcting the image's geometric distortion, and finally obtain a standard hyperspectral corrected image with accurate spatial location, no geometric distortion, and matching the contours of actual vegetation areas, providing a reliable foundation for subsequent image segmentation and canopy pixel extraction.

[0067] Preferably, the quantitative analysis module consists of a built-in industrial-grade embedded control board, a statistical computing chip, and a data comparison and storage unit. Relying on the built-in computing power, it invokes Pearson correlation logic to perform hardware-level correlation quantification calculations on high-dimensional spectral features and measured leaf water potential, soil moisture, and other growth environment data. It automatically calculates correlation coefficients and significance P-values, and uses hardware logic circuits to filter and remove weakly correlated and invalid features. The hardware solidifies the feature filtering rules and outputs a simplified and standardized spectral feature training set, achieving hardware-level purification from massive spectral features to an effective modeling dataset. The model training module, equipped with a built-in edge computing processor, machine learning hardware acceleration core, and model solidification storage chip, serves as the model training carrier. It imports the spectral feature training set and ground-measured ground-based ground-value data at the hardware level. Relying on the hardware acceleration unit, it builds a random forest network architecture and completes decision tree splitting, feature weight iteration, and nonlinear mapping relationship fitting through hardware parallel computing. The hardware solidifies the converged vegetation water parameter inversion model and stores it locally, enabling offline model training, parameter solidification, and local deployment, allowing for independent modeling capabilities even without cloud access. The parameter inversion module, supported by a dedicated system signal input interface, model calling hardware unit, and regression calculation chip, receives real-time hyperspectral reflectance information of target vegetation in the field. It automatically extracts target spectral features suitable for the model, calls the locally fixed inversion model, and rapidly inverts and outputs vegetation growth environment data such as leaf water potential and soil moisture through recursive calculations and multi-decision tree integration. This enables real-time hardware-level conversion of spectral information to water physiological and environmental parameters. The results analysis module integrates a built-in threshold storage chip, logic discrimination hardware circuit, data analysis processor, and display output interface. The hardware pre-stores critical thresholds for soil moisture and leaf water potential stress. At the hardware level, threshold comparison analysis is performed on soil moisture and leaf water potential information to generate two types of water stress information: environmental and physiological. Hardware logic circuits then complete a comprehensive evaluation of the two indicators and classify the stress level. Finally, standardized water stress diagnostic results are output via a physical display screen, data serial port, or IoT hardware port.

[0068] Another embodiment of the present invention provides a hyperspectral vegetation water stress diagnostic device; for details, please refer to [link to relevant documentation]. Figure 7 , Figure 7 The diagram shown illustrates a structural block diagram of a hyperspectral vegetation water stress diagnostic device according to one embodiment of the present invention. It includes a processor 21, a memory 22, and a computer program stored in the memory 22 and configured to be executed by the processor 21. When the processor 21 executes the computer program, it implements the steps described in the above embodiment of the hyperspectral vegetation water stress diagnostic method, for example... Figure 1 Steps S1 to S6 as described above.

[0069] For example, the computer program may be divided into one or more modules, which are stored in the memory 22 and executed by the processor 21 to complete the present invention. The one or more modules may be a series of computer program instruction segments capable of performing specific functions, which describe the execution process of the computer program in the hyperspectral vegetation water stress diagnostic device.

[0070] The hyperspectral vegetation water stress diagnostic device may include, but is not limited to, a processor 21 and a memory 22. Those skilled in the art will understand that the schematic diagram is merely an example of a hyperspectral vegetation water stress diagnostic device and does not constitute a limitation on such a device. It may include more or fewer components than illustrated, or combine certain components, or use different components. For example, the hyperspectral vegetation water stress diagnostic device may also include input / output devices, network access devices, buses, etc.

[0071] The processor 21 can be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor. The processor 21 is the control center of the hyperspectral vegetation water stress diagnostic device, connecting all parts of the device via various interfaces and lines.

[0072] The memory 22 can be used to store the computer program and / or modules. The processor 21 implements various functions of the hyperspectral vegetation water stress diagnostic device by running or executing the computer program and / or modules stored in the memory 22 and calling the data stored in the memory 22. The memory 22 may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the mobile phone (such as audio data, phonebook, etc.). In addition, the memory 22 may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0073] The module integrated into the hyperspectral vegetation water stress diagnostic device, if implemented as a software functional unit and sold or used as an independent product, can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium, etc.

[0074] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0075] Accordingly, embodiments of the present invention provide a computer-readable storage medium comprising a stored computer program, wherein, when the computer program is executed, it controls the device containing the computer-readable storage medium to perform steps in a hyperspectral vegetation water stress diagnosis method as described in the above embodiments, for example... Figure 1 Steps S1 to S6 as described above.

[0076] Compared to existing technologies, the beneficial effects of this invention are at least one of the following: This invention obtains a high-dimensional spectral feature set by acquiring hyperspectral image data of selected vegetation and extracting hyperspectral reflectance information for feature analysis, thus fully mining and utilizing the rich information of hyperspectral data. This solves the technical problem of existing technologies over-relying on a few preset vegetation indices, leading to the waste of a large number of sensitive spectral features. This invention quantifies the correlation between each high-dimensional spectral feature and actual growth environment data, and constructs a spectral feature training set based on the quantification results, achieving adaptive selection of features most relevant to water stress. This invention trains a vegetation water parameter inversion model based on the spectral feature training set and actual growth environment data, achieving automated, non-contact inversion of vegetation growth environment data, solving the problems of cumbersome operation and low efficiency of existing physical measurement methods. This invention addresses the technical challenges of meeting high-throughput or large-scale monitoring requirements. By inputting the hyperspectral reflectance information of the target vegetation into the inversion model, it obtains growth environment data including soil moisture and leaf water potential information, achieving a synergistic inversion of vegetation physiological state and environmental water conditions. This solves the problem of existing technologies only inverting a single water parameter and lacking sufficient information dimensions. Furthermore, by analyzing soil moisture and leaf water potential information, this invention obtains water stress results, enabling stress level diagnosis based on dual-index quantitative thresholds. This solves the technical problems of existing technologies lacking unified quantitative standards, having highly subjective diagnostic results, and varying from person to person. Finally, this invention achieves unified diagnosis of multiple water stress types, including drought and waterlogging stress, solving the technical problems of existing technologies focusing primarily on single drought stress and lacking systematic diagnostic capabilities for complex stresses such as waterlogging and rapid drought-waterlogging transitions.

[0077] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

Claims

1. A hyperspectral method for diagnosing vegetation water stress, characterized in that, include: Acquire hyperspectral image data of the selected vegetation and the actual growth environment data of the selected vegetation; Hyperspectral reflectance information is extracted from the hyperspectral image data, and feature analysis is performed on the hyperspectral reflectance information to obtain a high-dimensional spectral feature set; The correlation between each high-dimensional spectral feature in the high-dimensional spectral feature set and the actual growth environment data is quantified, and a spectral feature training set is constructed based on the quantization results; Based on the spectral feature training set and the actual growth environment data, the pre-constructed vegetation water parameter inversion model is trained to obtain the trained vegetation water parameter inversion model. The hyperspectral reflectance information of the target vegetation is input into the vegetation moisture parameter inversion model to obtain the growth environment data of the target vegetation. At least the soil moisture information and leaf water potential information in the growth environment data are analyzed to obtain the water stress results of the target vegetation.

2. The hyperspectral vegetation water stress diagnosis method as described in claim 1, characterized in that, The step of extracting hyperspectral reflectance information from the hyperspectral image data includes: The hyperspectral image data is subjected to radiometric calibration to obtain the image data to be corrected; Geometric correction processing is performed on the image data to be corrected to obtain standard hyperspectral image data; Pixel radiance data is extracted from the standard hyperspectral image data, and the pixel radiance data is analyzed to obtain the hyperspectral reflectance information.

3. The hyperspectral vegetation water stress diagnosis method as described in claim 1, characterized in that, The feature analysis of the hyperspectral reflectance information to obtain a high-dimensional spectral feature set includes: The hyperspectral reflectance information is denoised to obtain the denoised hyperspectral reflectance information; The denoised hyperspectral reflectance information is subjected to a first transformation process to obtain first-order spectral features; The denoised hyperspectral reflectance information is subjected to a second transformation process to obtain dual-band spectral features; The first-order spectral features and the dual-band spectral features are integrated to obtain the high-dimensional spectral feature set.

4. The hyperspectral method for diagnosing vegetation water stress as described in claim 1, characterized in that, The process of quantifying the correlation between each high-dimensional spectral feature in the high-dimensional spectral feature set and the actual growth environment data, and constructing a spectral feature training set based on the quantization results, includes: Based on the Pearson correlation principle, correlation analysis is performed on each of the high-dimensional spectral features and the actual growth environment data to obtain correlation information. The correlation information is quantified, and the high-dimensional spectral feature set is filtered based on the quantification results to determine the spectral feature training set.

5. The hyperspectral method for diagnosing vegetation water stress as described in claim 1, characterized in that, The process of training a pre-constructed vegetation moisture parameter inversion model based on the spectral feature training set and the actual growth environment data to obtain the trained vegetation moisture parameter inversion model includes: The spectral feature training set and the actual growth environment data are input into the data decision layer of the vegetation water parameter inversion model for analysis to obtain a nonlinear mapping relationship. Based on the nonlinear mapping relationship, the vegetation moisture parameter inversion model is trained to obtain the trained vegetation moisture parameter inversion model.

6. The hyperspectral method for diagnosing vegetation water stress as described in claim 1, characterized in that, The step of inputting the hyperspectral reflectance information of the target vegetation into the vegetation moisture parameter inversion model to obtain the growth environment data of the target vegetation includes: Extract the target spectral features from the hyperspectral reflectance information; Based on the decision tree in the vegetation moisture parameter inversion model, the target spectral features are recursively analyzed to obtain the growth environment data of the target vegetation.

7. The hyperspectral vegetation water stress diagnosis method as described in claim 1, characterized in that, The process of analyzing at least the soil moisture information and leaf water potential information in the growth environment data to obtain the water stress results of the target vegetation includes: Extract the soil moisture information and leaf water potential information from the growth environment data; Based on the soil moisture threshold, the soil moisture information is analyzed to obtain the first water stress information; Based on the leaf water potential threshold, the leaf water potential information is analyzed to obtain the second water stress information; The first water stress information and the second water stress information are comprehensively evaluated and processed to obtain the water stress result of the target vegetation.

8. The hyperspectral method for diagnosing vegetation water stress as described in claim 1, characterized in that, The acquisition of hyperspectral image data of the selected vegetation includes: The hyperspectral video data acquired by the hyperspectral imaging device is analyzed to obtain the first hyperspectral image frame data; Based on the regional information of the selected vegetation, the first hyperspectral image frame data is segmented to obtain the second hyperspectral image frame data. The second hyperspectral image frame data is subjected to environmental filtering to obtain the hyperspectral image data.

9. The hyperspectral method for diagnosing vegetation water stress as described in claim 1, characterized in that, After obtaining the water stress results, the method further includes: Based on the water stress results, the water stress level of the target vegetation is determined; Based on the water stress level, determine the maintenance information for the target vegetation; The maintenance information is converted into agricultural machinery control commands; The agricultural machinery control commands are executed to maintain the target vegetation.

10. A hyperspectral vegetation water stress diagnostic system, characterized in that, include: The data acquisition module is used to acquire hyperspectral image data of the selected vegetation and the actual growth environment data of the selected vegetation; The information extraction module is used to extract hyperspectral reflectance information from the hyperspectral image data and perform feature analysis on the hyperspectral reflectance information to obtain a high-dimensional spectral feature set. The quantification analysis module is used to quantify the correlation between each high-dimensional spectral feature in the high-dimensional spectral feature set and the actual growth environment data, and to construct a spectral feature training set based on the quantification results; The model training module is used to train the pre-constructed vegetation water parameter inversion model based on the spectral feature training set and the actual growth environment data, so as to obtain the trained vegetation water parameter inversion model. The parameter inversion module is used to input the hyperspectral reflectance information of the target vegetation into the vegetation water parameter inversion model to obtain the growth environment data of the target vegetation. The results analysis module is used to analyze at least the soil moisture information and leaf water potential information in the growth environment data to obtain the water stress results of the target vegetation.

11. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, wherein when the device containing the computer-readable storage medium executes the computer program, it implements a hyperspectral vegetation water stress diagnosis method as described in any one of claims 1 to 9.

12. A hyperspectral diagnostic device for vegetation water stress, characterized in that, The method includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor, when executing the computer program, implements a hyperspectral vegetation water stress diagnosis method as described in any one of claims 1 to 9.

Images