An image recognition-based water stress diagnosis method and system

Abstract

The application discloses a kind of water stress diagnosis method and system based on image recognition, it is related to the field of wisdom agriculture, including: acquisition module, for real-time acquisition crop canopy image, synchronous capture field illumination intensity, spectral distribution and illumination angle data, and complete image and illumination parameter binding storage according to preset frequency;Calibration module is used to receive binding data in acquisition module, establish pixel-level illumination compensation model by illumination angle partition, implement differentiating pixel brightness correction to different illumination areas;The application can adapt to complex illumination conditions in field, automatically correct image light and shadow distortion interference, accurately distinguish crops and background and stratified extraction crop characteristics, intelligently identify crop growth stage, accurately quantize and determine crop water stress grade, can also automatically generate standard diagnosis report, support data networking transmission and stress state visual display.

Description

Technical Field

[0001] This invention relates to the field of smart agriculture technology, specifically to a water stress diagnosis method and system based on image recognition. Background Technology

[0002] In agricultural production, crop water stress directly affects growth, development, and yield. Accurate diagnosis of its condition is key to scientific irrigation, improving water resource utilization, and ensuring food security. Traditional diagnostic methods are cumbersome, time-consuming, and destructive, making it difficult to achieve large-scale real-time monitoring. However, the development of image recognition and artificial intelligence technologies has provided a new path for efficient diagnosis.

[0003] The invention patent application with application number 201610465992.4 discloses a method for detecting water stress phenotypes in crop plants based on image processing. This application aims to solve the problem that "current plant phenotype detection methods require a certain degree of manual interaction, are not sufficiently automated, cannot accurately perceive and monitor subtle changes in crops, and cannot accurately reflect the interaction between crops and the environment".

[0004] However, existing image recognition-based methods for diagnosing water stress are easily affected by environmental factors such as field lighting and cluttered backgrounds, and they are difficult to take into account the phenotypic differences of crops at different growth stages, resulting in insufficient diagnostic accuracy and environmental adaptability.

[0005] To address this, we propose an image recognition-based method and system for diagnosing water stress. Summary of the Invention

[0006] In view of the above-mentioned shortcomings of the existing technology, the present invention provides a water stress diagnosis method and system based on image recognition, which can effectively solve the problems of the existing technology.

[0007] To achieve the above objectives, the present invention is implemented through the following technical solutions; This invention discloses a water stress diagnosis method and system based on image recognition, comprising: The system comprises the following modules: Acquisition module, Calibration module, and Output module. Acquisition module acquires crop canopy images in real time, simultaneously capturing field light intensity, spectral distribution, and light angle data, and binding and storing images with light parameters at a preset frequency. Calibration module receives the bound data from the acquisition module, establishes a pixel-level light compensation model through light angle partitioning, and performs differentiated pixel brightness correction for different light areas. Extraction module receives the corrected images from the calibration module, divides the crop area into crop and background areas, extracts morphological, color, and texture features from different crop organs, and filters background interference features. Matching module receives the extracted features from the extraction module, compares them with a preset standard feature library for each crop growth stage, identifies the current crop growth stage, and calls the corresponding water stress diagnostic parameter set. Diagnosis module quantifies the extracted features based on the growth stage identified by the matching module and the called diagnostic parameter set, establishes a mapping relationship between feature parameters and the degree of water stress, and quantitatively determines the water stress state. Output module receives the quantitative determination results from the diagnosis module, generates a standardized diagnostic report in a preset format, and supports data interface output and visualization. The acquisition module is interconnected with a calibration module and an extraction module via a wireless network. The extraction module is interconnected with the calibration module via a wireless network. The extraction module is interconnected with a matching module via a wireless network. The matching module is interconnected with a diagnostic module and an output module via a wireless network. The output module and the diagnostic module are interconnected via a wireless network.

[0008] Furthermore, the acquisition module is equipped with a multi-source synchronous sensing unit, which integrates an image acquisition unit, a light intensity sensing unit, a spectrum acquisition unit, and an angle detection unit. While the image acquisition unit continuously acquires original images of the crop canopy, the light intensity sensing unit, the spectrum acquisition unit, and the angle detection unit instantaneously and synchronously capture the corresponding field light parameters. The image acquisition time is used as the time anchor point to establish a timestamp association. The preset frequency is set according to the system's built-in time sequence triggering rules. The canopy images and multi-dimensional light parameters under the same timestamp are encapsulated into a bound data package and stored in a dedicated cache storage partition.

[0009] Furthermore, the calibration module divides the field illumination angle into zones based on the spatial projection range of the field illumination. It constructs a pixel-level illumination compensation model based on the light propagation attenuation law within each zone. For each illumination angle zone, it independently fits a pixel brightness reference response curve. Using the zone reference response curve as the correction reference, it performs pixel-by-pixel differential brightness offset correction on each pixel in the canopy image to eliminate image pixel distortion interference caused by differences in illumination angle, light intensity, and spectral distribution.

[0010] Furthermore, the extraction module defines the crop region and background region based on the image semantic edge segmentation algorithm, completes the organ hierarchy division according to the crop root-crown organ growth structure, and independently collects the original features of morphology, color, and texture of each organ layer by layer to construct an organ feature coupling representation model to calculate the effective feature values ​​of each layer, and removes redundant background feature information mixed into the crop region. The organ feature coupling representation model satisfies the following formula: ; In the formula: This represents the effective characteristic characterization value for single crop organ stratification; Dimensionless values ​​normalized to organ morphological characteristics; This refers to the spectral offset of organ color characteristics; Entropy of the spatial distribution of organ texture features; This is the background feature interference offset correction amount; After obtaining the effective feature representation value S of a single crop organ layer, the extraction module has a built-in preset feature discrimination threshold. It calculates the corresponding feature representation value S for each pixel unit in the crop area, compares the representation value of each pixel unit with the preset feature discrimination threshold, and identifies the pixel unit whose representation value is lower than the preset feature discrimination threshold as background redundant feature pixels mixed into the crop area. The original features of the shape, color and texture of such pixels are masked and cleared by the secondary masking method of the region mask, thus completing the targeted filtering and removal of background redundant feature information.

[0011] Furthermore, the matching module pre-stores original feature samples of the entire crop growth stage and constructs a standard feature library for each growth stage through sample clustering iteration. The layered features output by the extraction module are compared with the feature samples of each growth stage in the standard feature library in terms of dimensional spatial similarity. When the feature matching degree falls into the preset matching interval, the current growth stage of the crop is determined, and the water stress diagnosis parameter set specific to that growth stage is automatically retrieved to lock the appropriate stress judgment rule.

[0012] Furthermore, the diagnostic module transforms the layered extracted morphological, color, and texture features into a quantifiable sequence of feature parameters. Based on the crop's physiological water shortage response mechanism, it establishes a nonlinear correlation mapping framework between the feature parameter sequence and the water stress gradient. The diagnostic module analyzes the evolution of each feature parameter with varying water stress conditions based on this mapping framework, and, combined with the retrieved diagnostic parameter set, delineates the stress level determination boundary to perform a continuous quantitative determination of the degree of crop water stress. ; In the formula: G is the uninterrupted quantitative level characterization value of crop water stress; H is the comprehensive feature mapping quantity after the feature parameter sequence is transformed by the nonlinear mapping framework; K is the diagnostic benchmark correction coefficient corresponding to the current crop growth stage; Q is the gradient offset of field environmental disturbance on water stress determination.

[0013] Furthermore, the output module integrates quantitative water stress determination results, crop growth stage information, original canopy images, and light-related parameters to generate a standardized diagnostic report according to a preset report architecture. It also supports adapting diagnostic data to general industrial data interfaces and agricultural IoT transmission interfaces, and displays the real-time status and evolution trend of crop water stress in a visualized form of hierarchical dimensional maps and time-series change curves.

[0014] On the other hand, an image recognition-based method for diagnosing water stress includes: The system acquires crop canopy images in real time, simultaneously obtaining data on field light intensity, spectral distribution, and light angle. Images are then bound to light parameters at preset frequencies and stored. Regions are divided based on light angle, and a pixel-level light compensation model is constructed to perform differentiated brightness correction on each pixel of the canopy image, eliminating image distortion interference caused by light differences. Image semantic edge segmentation is used to separate crop and background regions, and morphological, color, and texture features of various crop organs are extracted layer by layer. Redundant background interference features are filtered through an organ feature coupling representation model. The extracted crop features are compared with standard feature libraries for each growth stage to identify the current crop growth stage and call the corresponding water stress diagnostic parameter set and stress judgment rules. Crop feature parameters are quantitatively analyzed, and a nonlinear mapping relationship between feature parameters and water stress gradient is established. Combined with the diagnostic parameter set, the degree of water stress is continuously and quantitatively determined. The water stress judgment results, crop growth stage, canopy images, and light parameters are integrated to generate a standardized diagnostic report, supporting transmission through multiple data interfaces and visualized in the form of graphs and time-series curves.

[0015] Compared with the known prior art, the technical solution provided by this invention has the following beneficial effects: This invention simultaneously acquires and stores crop canopy images and field illumination parameters. Based on pixel-level illumination compensation, it corrects image distortion and avoids imaging interference caused by complex field illumination. It accurately divides crop and background areas using semantic segmentation, extracts crop morphology, color, and texture features layer by layer, and filters out redundant background information. Based on a growth stage feature library, it accurately identifies the crop growth period, matches specific diagnostic parameters and judgment rules, establishes a correlation mapping between phenotypic features and water stress, and achieves quantitative and graded judgment of stress status. It can also automatically generate standardized diagnostic reports, adapt to agricultural data interface transmission, and intuitively present the real-time status and evolution trend of stress in a visual manner, effectively improving the accuracy, practicality, and convenience of crop water stress diagnosis. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0017] Figure 1 This is a schematic diagram of a water stress diagnosis system based on image recognition. Figure 2 This is a flowchart illustrating a water stress diagnosis method based on image recognition. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0019] The present invention will be further described below with reference to embodiments.

[0020] Example 1: This embodiment presents a water stress diagnosis system based on image recognition, such as... Figure 1 As shown, it includes: The acquisition module is used to acquire crop canopy images in real time, simultaneously capture field light intensity, spectral distribution and light angle data, and complete the binding and storage of images and light parameters at a preset frequency; The acquisition module is equipped with a multi-source synchronous sensing unit, which integrates an image acquisition unit, a light intensity sensing unit, a spectrum acquisition unit, and an angle detection unit. While the image acquisition unit continuously acquires original images of the crop canopy, the light intensity sensing unit, the spectrum acquisition unit, and the angle detection unit instantaneously and synchronously capture the corresponding field light parameters. The image acquisition time is used as the time anchor point to establish a timestamp association. The preset frequency is set according to the system's built-in time sequence triggering rules. The canopy images and multi-dimensional light parameters under the same timestamp are encapsulated into a bound data package and stored in a dedicated cache storage partition. The preset frequency setting ranges from 1 to 30 frames per minute and can be manually or adaptively adjusted according to the density of crop canopy growth and the rate of change in field light intensity. The system's built-in time-series triggering rules adopt dual modes of equal time interval triggering and light change triggering. The equal time interval is sampled at a fixed time sequence according to the preset frequency. When the instantaneous fluctuation of light intensity exceeds ±10%, additional sampling is automatically triggered. All sampling moments generate a unique timestamp, and the canopy image and multi-dimensional light parameters are linked with millisecond-level time precision anchor points. The calibration module receives data bound to the acquisition module, establishes a pixel-level illumination compensation model through illumination angle partitioning, and performs differentiated pixel brightness correction for different illumination areas. The calibration module divides the field light space projection range into light angle zones, constructs a pixel-level light compensation model based on the light propagation attenuation law in each zone, independently fits the pixel brightness reference response curve for each light angle zone, and uses the zone reference response curve as the correction reference to perform pixel-by-pixel differential brightness shift correction on each pixel in the canopy image to eliminate image pixel distortion interference caused by differences in light angle, light intensity and spectral distribution. Among them, when dividing the field light space projection range as the dividing benchmark to complete the light angle partitioning, the field space projection boundary corresponding to the natural light incident direction and height is based on the crop canopy three-dimensional geometric contour to adaptively delineate the light coverage area, autonomously distinguish the direct light area, diffuse light area and canopy shadow shading area, and form an angle partitioning with the boundary dynamically changing with the crop growth in the field. The initial division of the illumination angle partitioning is completed with an azimuth angle of 15° as the step size and an elevation angle of 10° as the interval benchmark. The threshold of the change in the average gray value of the canopy pixels is used as the boundary judgment standard for the direct light area, the diffuse light area, and the canopy shadow occlusion area. The crop growth is judged by the growth rate of the canopy projection area. When the change in the canopy projection area exceeds 5% in three consecutive samplings, the illumination space projection boundary is automatically refitted, the illumination angle partitioning range is updated, and the partitioning boundary is dynamically and adaptively adjusted. In the stage of constructing a pixel-level illumination compensation model based on the light propagation attenuation law in each zone, the characteristics of light energy attenuation layer by layer caused by field atmospheric scattering and canopy foliage shading are combined to fit the continuous correlation between pixel gray values ​​and light angle changes in each zone, establish the illumination attenuation response relationship for each pixel position in the image, and then construct a pixel-level exclusive illumination compensation model covering the entire canopy image. When performing pixel-by-pixel differential brightness offset correction on each pixel in the canopy image, the corresponding illumination attenuation response relationship is matched according to the illumination angle of the pixel, the inherent brightness deviation amplitude of a single pixel is solved, and a dedicated correction offset is matched according to the spatial position of the pixel and the illumination characteristics of the partition. Each pixel independently completes brightness correction using correction parameters adapted to its own working conditions, without a globally unified fixed correction parameter. The extraction module is used to receive the corrected image from the calibration module, divide the crop area and the background area, extract morphological, color and texture features of different organs of the crop in layers, and filter background interference feature information. The extraction module defines the crop region and background region based on the image semantic edge segmentation algorithm, completes the organ hierarchy division according to the crop root-crown organ growth structure, and independently collects the original features of morphology, color, and texture of each organ layer to construct an organ feature coupling representation model to calculate the effective feature values ​​of each layer, and removes redundant background feature information mixed into the crop region. The organ feature coupling representation model satisfies the following formula: ; In the formula: This represents the effective characteristic value for crop single organ stratification; It is a dimensionless value normalized to the morphological characteristics of organs, representing the morphological attributes of organ outline, plant type and canopy density; This refers to the spectral offset of organ color features, characterizing the degree of color shift of crop canopy pixels relative to a healthy baseline color. Entropy represents the spatial distribution of organ texture features, characterizing the disordered distribution characteristics of crop surface texture arrangement; This is the background feature interference offset correction amount, which represents the magnitude of the interference offset of background clutter pixels on crop organ feature extraction; in, The maximum-minimum normalization algorithm is used to uniformly map organ contour curvature, canopy projection ratio, and plant extension scale to the dimensionless interval [0,1]. By extracting the pixel components of RGB and multispectral channels, the Euclidean distance between the actual spectral value of crop organs and the healthy baseline spectral value is calculated as the spectral offset. The probability of grayscale distribution of local texture in an organ is statistically analyzed using the classical formula for calculating information entropy, and the spatial distribution entropy is then solved. The absolute deviation magnitude between the mean value of background clutter pixel features and the mean value of standard crop organ features is calculated and used as the background feature interference offset correction amount. All parameters are quantized and uniformly substituted into the coupled model for calculation. The above formula integrates the three core phenotypic features of crop organs—morphology, color, and texture—for coupled calculation. It uses the normalized value of morphological features as the basic weight, weakens the interference caused by extreme offset values ​​of color features by taking the square root, and constructs the calculation denominator by subtracting the background interference correction amount from the spatial distribution entropy of texture features. It achieves dimensionless quantification of organ features from multiple dimensions, and at the same time, it specifically offsets the feature extraction deviation caused by background clutter pixels in the field. It accurately calculates the effective feature values ​​of crop single organ layer, and is suitable for the actual scenario of accurate extraction of crop organ features in complex field backgrounds. Among them, the image semantic edge segmentation algorithm is configured with multi-scale feature perception channels and preset semantic label set, using shallow texture edges and deep semantic features of canopy images as dual judgment criteria, and setting preset multi-scale perception range to adapt to the density variation characteristics of crop canopy. The region delineation process is as follows: First, the calibrated complete canopy image is input into the algorithm model. The gradient edge features and deep semantic association features of the image are extracted simultaneously through the multi-scale feature perception channel. Then, the semantic attributes of the whole image pixels are classified pixel by pixel in combination with the preset semantic label set. Subsequently, connected component aggregation and clustering are performed on the semantic pixels of the same type. Effective connected components are selected according to the prior constraints of crop growth morphology. Finally, a binary region mask is generated to complete the closure morphology division of the crop entity region and the surrounding background region. After obtaining the effective feature representation value S of a single crop organ layer, the extraction module has a built-in preset feature discrimination threshold. It calculates the corresponding feature representation value S for each pixel unit in the crop area, compares the representation value of each pixel unit with the preset feature discrimination threshold, and identifies the pixel unit whose representation value is lower than the preset feature discrimination threshold as background redundant feature pixels mixed into the crop area. The original features of the shape, color and texture of such pixels are masked and cleared by the secondary masking method of the region mask, thus completing the targeted filtering and removal of background redundant feature information. The feature discrimination threshold was determined through field calibration experiments on multiple varieties and all growth stages, with the threshold value range set between 0.2 and 0.6. The system has a built-in threshold mapping table that associates crop varieties with growth stages, which can automatically retrieve the corresponding threshold based on the crop growth stage identified by the matching module. When the proportion of background debris in the field exceeds 15%, the threshold is adaptively and slightly increased in a linear proportion to ensure the accuracy of background redundant feature filtering while retaining the effective feature pixels of the crop. in, Quantification was achieved by performing dimensionless normalization fusion calculations on crop organ contour curvature, canopy projection ratio, and plant extension scale. Quantization is performed by extracting the offset mapping relationship between the organ pixel spectral components and the crop health baseline spectral components; Quantification is achieved by statistically analyzing the probability of grayscale distribution of local texture in organs and solving for spatial information entropy. Quantization is performed by calculating the magnitude of the deviation between the mean value of background pixel features and the mean value of standard organ features of crops; The matching module is used to receive the extracted features output by the extraction module, compare them with the preset standard feature library of each crop growth stage, identify the current crop growth stage, and call the water stress diagnosis parameter set of the corresponding stage. The matching module pre-stores the original feature samples of the entire crop growth stage, and constructs a standard feature library for each growth stage through sample clustering iteration; The K-Means clustering algorithm is used to carry out iterative clustering of original feature samples. The number of initial clusters is set according to the number of crop growth stages. The convergence threshold of the cluster center iteration offset is set to be less than 0.01. The clustering is considered to have converged if the cluster center offset is lower than this threshold for two consecutive rounds. The high-dimensional feature space similarity is calculated using the cosine similarity algorithm. The cosine value of the angle between the feature vectors is used as the matching metric. The matching degree interval [0.85, 1] ​​is used to determine the matching of the corresponding growth stage. The layered features output by the extraction module are compared with the feature samples of each growth stage in the standard feature library in terms of dimensional space similarity. When the feature matching degree falls into the preset matching interval, the current growth stage of the crop is determined, and the water stress diagnosis parameter set specific to that growth stage is automatically retrieved to lock the appropriate stress judgment rule. During the extraction module's operation phase, the pre-stored original feature samples of the entire crop growth stage are first pre-processed with multi-dimensional feature normalization. The initial category division is completed using the natural growth sequence of the crop as the clustering constraint. Then, the cluster center iterative offset is used as the convergence criterion. The cluster center is iteratively updated round by round, and discrete abnormal feature samples that deviate from the cluster distribution are automatically removed until the cluster center offset amplitude converges to the preset fluctuation threshold. Each cluster sample after convergence is then solidified to correspond to a single crop growth stage, thereby completing the construction of the standard feature library for each growth stage. When performing dimensional space similarity comparison, the crop organ hierarchical features output by the extraction module are mapped to a high-dimensional feature space with the same dimension as the standard feature library. The feature matching degree is calculated by jointly quantizing the spatial orientation deviation and vector magnitude difference of the high-dimensional feature vectors, thereby completing accurate similarity comparison without dimensionality reduction loss. Among them, the water stress diagnostic parameter set is pre-calibrated, classified, packaged and stored by the system based on the controlled cultivation experiment of crop growth at each growth stage with gradient water, and the corresponding crop phenotypic characteristics are associated with the response change law of water gradient. The adapted stress determination rules are jointly customized based on the physiological water shortage critical characteristics and phenotypic stress response boundary characteristics of crops at the corresponding growth stage, and are bound and stored one by one with the water stress diagnostic parameter set at the corresponding growth stage. When the diagnostic parameter set is retrieved, the matching stress determination rules are loaded simultaneously. The diagnostic module is used to perform quantitative analysis on the extracted features based on the reproductive stage identified by the matching module and the set of diagnostic parameters invoked, to establish a mapping relationship between the feature parameters and the degree of water stress, and to complete the quantitative determination of the water stress state. The diagnostic module transforms the morphological, color, and texture features extracted layer by layer into a quantifiable sequence of feature parameters. Based on the physiological water shortage response mechanism of crops, it establishes a nonlinear correlation mapping framework between the feature parameter sequence and the water stress gradient. The diagnostic module analyzes the evolution of each feature parameter with the changing water stress state based on this mapping framework, and, combined with the retrieved diagnostic parameter set, delineates the stress level determination boundary to perform a continuous quantitative determination of the degree of crop water stress. ; In the formula: G is the uninterrupted quantitative level characterization value of crop water stress; H is the comprehensive feature mapping amount after the feature parameter sequence is transformed by the nonlinear mapping framework; K is the diagnostic benchmark correction coefficient corresponding to the current crop growth stage; Q is the gradient offset of field environmental disturbance on water stress judgment; The above formula combines the natural logarithm and the exponential function to build a nonlinear computational architecture, incorporating the comprehensive feature mapping quantity, the crop growth stage diagnostic benchmark correction coefficient, and the field environmental disturbance gradient offset. It uses the exponential function to simulate the gradual influence of environmental disturbance, and the logarithmic function to smooth the numerical fluctuation of water stress level. At the same time, it introduces correction coefficients for adaptation and calibration based on the physiological water requirement characteristics of crops at different growth stages. This breaks the traditional discrete level classification mode and can continuously and uninterruptedly quantify the degree of crop water stress, which meets the actual needs of refined judgment of crop water stress in complex field environments. K is obtained by coupling and normalizing the deviation of the health phenotypic baseline characteristics of crops at each growth stage with the difference in physiological water requirement threshold. Q takes field temperature, humidity, and light intensity fluctuations as input variables and solves the environmental disturbance gradient offset through a multivariate linear fitting model. The nonlinear correlation mapping framework adopts a three-layer BP neural network architecture, taking morphological, color, and texture feature parameters as inputs and water stress gradient labels as outputs. After training, the network weights are solidified to realize the nonlinear mapping between features and stress gradients. The controlled cultivation experiment set up four water and fertilizer gradients: severe stress, moderate stress, mild stress, and normal water supply. The crop water stress level was divided into five levels, corresponding to no stress, mild stress, moderate stress, severe stress, and extreme stress. The gradient experiment was used to fit the correspondence between the characteristic parameters of each growth stage and the stress level, and the upper and lower boundaries of the characteristic parameters of each stress level were defined and encapsulated to form a unique water stress diagnostic parameter set. The output module is used to receive the quantitative judgment results output by the diagnostic module, generate standardized diagnostic reports in a preset format, and support data interface output and visualization. The standardized diagnostic report architecture includes six core sections: collection time, crop growth stage, light parameters, feature extraction data, water stress level, and irrigation recommendations. The data interface is compatible with the Modbus industrial protocol and the MQTT agricultural IoT transmission protocol, supporting encrypted transmission and timed breakpoint resume. The visualized hierarchical dimension map displays the feature distribution by crop organ, and the time-series change curve is sampled at 5-minute intervals to continuously display the dynamic evolution trend of water stress level within 24 hours. The output module integrates quantitative water stress assessment results, crop growth stage information, original canopy images, and light-related parameters, generates standardized diagnostic reports according to a preset report architecture, and supports the adaptation of diagnostic data to general industrial data interfaces and agricultural IoT transmission interfaces. At the same time, it displays the real-time status and evolution trend of crop water stress in a visual form of hierarchical dimensional maps and time-series change curves. The acquisition module is interconnected with the calibration module and the extraction module via a wireless network. The extraction module is interconnected with the calibration module via a wireless network. The extraction module is interconnected with the matching module via a wireless network. The matching module is interconnected with the diagnostic module and the output module via a wireless network. The output module and the diagnostic module are interconnected via a wireless network.

[0021] In this embodiment, the acquisition module acquires crop canopy images in real time, simultaneously capturing data on field light intensity, spectral distribution, and light angle. It then binds and stores the images and light parameters at a preset frequency. The calibration module, running subsequently, receives the bound data from the acquisition module, establishes a pixel-level light compensation model through light angle partitioning, and performs differentiated pixel brightness correction for different light areas. The extraction module receives the corrected images from the calibration module in real time, divides the crop area into crop and background areas, extracts morphological, color, and texture features from different crop organs, and filters background interference features. The matching module receives the extracted features from the extraction module, compares them with a preset standard feature library for each crop growth stage, identifies the current crop growth stage, and calls the corresponding water stress diagnostic parameter set. The diagnosis module further quantifies the extracted features based on the growth stage identified by the matching module and the called diagnostic parameter set, establishing a mapping relationship between feature parameters and the degree of water stress and completing a quantitative determination of the water stress state. Finally, the output module receives the quantitative determination results output by the diagnosis module, generates a standardized diagnostic report in a preset format, and supports data interface output and visualization.

[0022] In the above embodiments, the system can adapt to complex field lighting conditions, automatically correct image light and shadow distortion interference, accurately distinguish between crops and background and extract crop features in layers, intelligently identify crop growth stages, accurately quantify and determine the crop water stress level, and automatically generate standard diagnostic reports. It also supports data network transmission and visualization of stress status, providing real-time and reliable technical data support for precision irrigation of farmland and refined crop growth management.

[0023] Example 2: At the implementation level, based on Example 1, this example refers to... Figure 2 A further detailed description of the water stress diagnosis system based on image recognition in Example 1 is provided below: A water stress diagnosis method based on image recognition, comprising: Real-time acquisition of crop canopy images, simultaneous acquisition of field light intensity, spectral distribution and light angle data, and storage of images after binding with light parameters at a preset frequency; The region is divided according to the illumination angle, a pixel-level illumination compensation model is constructed, and differential brightness correction is performed on each pixel of the canopy image to eliminate image distortion interference caused by illumination differences; Image semantic edge segmentation is used to divide the crop and background regions, and the morphological, color and texture features of each crop organ are extracted in layers. The redundant interference features of the background are filtered out by the organ feature coupling representation model. The extracted crop features are compared with the standard feature library for each growth stage to identify the current crop growth stage, and the corresponding water stress diagnosis parameter set and stress judgment rules are called. Quantitative analysis of crop characteristic parameters is performed to establish a nonlinear mapping relationship between characteristic parameters and water stress gradient, and the degree of water stress is continuously and quantitatively determined by combining diagnostic parameter set. It integrates water stress assessment results, crop growth stage, canopy images, and light parameters to generate standardized diagnostic reports, supports the transmission of multiple data interfaces, and provides visual displays in the form of graphs and time-series curves.

[0024] In summary, the system and method in the above embodiments simultaneously acquire and store crop canopy images and field illumination parameters, correct image distortion based on pixel-level illumination compensation, avoid imaging interference caused by complex field illumination, accurately divide crop and background areas using semantic segmentation, extract crop morphology, color, and texture features layer by layer and filter out redundant background information, accurately identify crop growth stages based on a growth stage feature library, match specific diagnostic parameters and judgment rules, establish a correlation mapping between phenotypic features and water stress, achieve quantitative and graded judgment of stress status, automatically generate standardized diagnostic reports, adapt to agricultural data interface transmission, and intuitively present the real-time status and evolution trend of stress in a visual manner, effectively improving the accuracy, practicality, and convenience of crop water stress diagnosis.

[0025] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions will not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A water stress diagnostic system based on image recognition, characterized in that, include: The acquisition module is used to acquire crop canopy images in real time, simultaneously capture field light intensity, spectral distribution and light angle data, and complete the binding and storage of images and light parameters at a preset frequency; The calibration module receives data bound to the acquisition module, establishes a pixel-level illumination compensation model through illumination angle partitioning, and performs differentiated pixel brightness correction for different illumination areas. The extraction module is used to receive the corrected image from the calibration module, divide the crop area and the background area, extract morphological, color and texture features of different organs of the crop in layers, and filter background interference feature information. The matching module is used to receive the extracted features output by the extraction module, compare them with the preset standard feature library of each crop growth stage, identify the current crop growth stage, and call the water stress diagnosis parameter set of the corresponding stage. The diagnostic module is used to perform quantitative analysis on the extracted features based on the reproductive stage identified by the matching module and the set of diagnostic parameters invoked, to establish a mapping relationship between the feature parameters and the degree of water stress, and to complete the quantitative determination of the water stress state. The output module receives the quantitative judgment results output by the diagnostic module, generates standardized diagnostic reports in a preset format, and supports data interface output and visualization.

2. The water stress diagnosis system based on image recognition according to claim 1, characterized in that, The acquisition module is equipped with a multi-source synchronous sensing unit, which integrates an image acquisition unit, a light intensity sensing unit, a spectrum acquisition unit, and an angle detection unit. While the image acquisition unit continuously acquires original images of the crop canopy, the light intensity sensing unit, the spectrum acquisition unit, and the angle detection unit instantaneously and synchronously capture the corresponding field light parameters. The image acquisition time is used as the time anchor point to establish a timestamp association. The preset frequency is set according to the system's built-in time sequence triggering rules. The canopy images and multi-dimensional light parameters under the same timestamp are encapsulated into a bound data package and stored in a dedicated cache storage partition.

3. The water stress diagnostic system based on image recognition according to claim 1, characterized in that, The calibration module divides the field illumination angle into zones based on the spatial projection range of the field illumination. It constructs a pixel-level illumination compensation model based on the light propagation attenuation law within each zone. It independently fits the pixel brightness reference response curve for each illumination angle zone. Using the zone reference response curve as the correction reference, it performs pixel-by-pixel differential brightness offset correction on each pixel in the canopy image to eliminate image pixel distortion interference caused by differences in illumination angle, light intensity, and spectral distribution.

4. The water stress diagnosis system based on image recognition according to claim 1, characterized in that, The extraction module defines the crop region and background region based on the image semantic edge segmentation algorithm, completes the organ hierarchy division according to the crop root-crown organ growth structure, and independently collects the original features of morphology, color, and texture of each organ layer by layer to construct an organ feature coupling representation model to calculate the effective feature values ​​of each layer, and removes redundant background feature information mixed into the crop region. The organ feature coupling representation model satisfies the following formula: ； In the formula: This represents the effective characteristic characterization value for single crop organ stratification; Dimensionless values ​​normalized to organ morphological characteristics; This refers to the spectral offset of organ color characteristics; Entropy of the spatial distribution of organ texture features; This is the background feature interference offset correction amount; After obtaining the effective feature representation value S of a single crop organ layer, the extraction module has a built-in preset feature discrimination threshold. It calculates the corresponding feature representation value S for each pixel unit in the crop area, compares the representation value of each pixel unit with the preset feature discrimination threshold, and identifies the pixel unit whose representation value is lower than the preset feature discrimination threshold as background redundant feature pixels mixed into the crop area. The original features of the shape, color and texture of such pixels are masked and cleared by the secondary masking method of the region mask, thus completing the targeted filtering and removal of background redundant feature information.

5. The water stress diagnostic system based on image recognition according to claim 1, characterized in that, The matching module pre-stores the original feature samples of the entire crop growth stage, and constructs a standard feature library for each growth stage through sample clustering iteration. The layered features output by the extraction module are compared with the feature samples of each growth stage in the standard feature library in terms of dimensional spatial similarity. When the feature matching degree falls into the preset matching interval, the current growth stage of the crop is determined, and the water stress diagnosis parameter set specific to that growth stage is automatically retrieved to lock the appropriate stress judgment rule.

6. The water stress diagnostic system based on image recognition according to claim 1, characterized in that, The diagnostic module transforms the morphological, color, and texture features extracted layer by layer into a quantifiable sequence of feature parameters. Based on the physiological water shortage response mechanism of crops, it establishes a nonlinear correlation mapping framework between the feature parameter sequence and the water stress gradient. The diagnostic module analyzes the evolution of each feature parameter with the water stress state based on the mapping framework, and delineates the stress level judgment boundary in combination with the retrieved diagnostic parameter set, thus performing a continuous quantitative judgment of the degree of crop water stress. ； In the formula: G is the uninterrupted quantitative level characterization value of crop water stress; H is the comprehensive feature mapping quantity after the feature parameter sequence is transformed by the nonlinear mapping framework; K is the diagnostic benchmark correction coefficient corresponding to the current crop growth stage; Q is the gradient offset of field environmental disturbance on water stress determination.

7. The water stress diagnosis system based on image recognition according to claim 1, characterized in that, The output module integrates quantitative water stress assessment results, crop growth stage information, original canopy images, and light-related parameters, generates standardized diagnostic reports according to a preset report architecture, and supports the adaptation of diagnostic data to general industrial data interfaces and agricultural IoT transmission interfaces. At the same time, it displays the real-time status and evolution trend of crop water stress in a visualized form of hierarchical dimensional maps and time-series change curves.

8. The water stress diagnosis system based on image recognition according to claim 1, characterized in that, The acquisition module is interconnected with a calibration module and an extraction module via a wireless network. The extraction module is interconnected with the calibration module via a wireless network. The extraction module is interconnected with a matching module via a wireless network. The matching module is interconnected with a diagnostic module and an output module via a wireless network. The output module and the diagnostic module are interconnected via a wireless network.

9. A water stress diagnosis method based on image recognition, wherein the method is an implementation method of the water stress diagnosis method and system based on image recognition as described in any one of claims 1-8, characterized in that, include: Real-time acquisition of crop canopy images, simultaneous acquisition of field light intensity, spectral distribution and light angle data, and storage of images after binding with light parameters at a preset frequency; The region is divided according to the illumination angle, a pixel-level illumination compensation model is constructed, and differential brightness correction is performed on each pixel of the canopy image to eliminate image distortion interference caused by illumination differences; Image semantic edge segmentation is used to divide the crop and background regions, and the morphological, color and texture features of each crop organ are extracted in layers. The redundant interference features of the background are filtered out by the organ feature coupling representation model. The extracted crop features are compared with the standard feature library for each growth stage to identify the current crop growth stage, and the corresponding water stress diagnosis parameter set and stress judgment rules are called. Quantitative analysis of crop characteristic parameters is performed to establish a nonlinear mapping relationship between characteristic parameters and water stress gradient, and the degree of water stress is continuously and quantitatively determined by combining diagnostic parameter set. It integrates water stress assessment results, crop growth stage, canopy images, and light parameters to generate standardized diagnostic reports, supports the transmission of multiple data interfaces, and provides visual displays in the form of graphs and time-series curves.

Images