A system and method for assessing the health status of tomato foliage

Abstract

The application provides a tomato leaf health state evaluation system and method, relates to the technical field of tomato leaf health evaluation, and specifically comprises the following steps: firstly, the tomato leaf image is preprocessed and converted to the Lab color space, and the leaf area is extracted based on the chroma threshold segmentation. Then, the chroma channel statistics of the leaf area are counted, pixel-level anomaly detection is carried out through an adaptive threshold, and abnormal and normal area masks are generated. The abnormal area is subjected to connected domain analysis, and local color and texture features are extracted, while the corresponding features of the normal area are extracted as references. The health deviation is obtained by calculating the feature distance between each abnormal area and the normal area, and the comprehensive health evaluation index is generated after normalization combined with the area proportion of the abnormal area. Finally, the health state is determined according to the preset threshold rule, and the evaluation result is output. The application provides comprehensive and accurate decision-making basis for agricultural producers, and improves the automation level and evaluation accuracy of tomato disease monitoring.

Description

Technical Field

[0001] This invention relates to the field of tomato leaf health assessment technology, specifically to a tomato leaf health status assessment system and method. Background Technology

[0002] In tomato cultivation, early identification and accurate assessment of foliar diseases are crucial for ensuring crop yield and quality. Currently, mainstream monitoring methods still rely on field inspections by agricultural personnel, which suffers from low efficiency, high subjectivity, and insensitivity to early-stage diseases. With the development of information technology, automated detection technologies based on computer vision have begun to be applied to plant health monitoring. However, most existing solutions employ end-to-end classification methods based on deep learning, which have significant limitations. These methods typically require large-scale labeled data for model training, resulting in high computational complexity and difficulty in deployment on resource-constrained edge devices. More importantly, existing solutions often analyze leaves as a whole, failing to effectively identify and quantify early localized lesions and lacking the ability to accurately assess disease severity.

[0003] Existing technical solutions face several key technical bottlenecks in practical implementation. First, most methods process directly in the RGB color space, making them sensitive to changes in lighting and unable to effectively separate leaves from complex backgrounds. Second, segmentation methods based on fixed thresholds have poor adaptability and cannot handle the natural color variations of leaves at different varieties and growth stages. Third, existing methods often focus on overall classification while neglecting local feature analysis, lacking sensitivity to detect minute anomalies. Furthermore, traditional methods typically utilize only color features while ignoring texture information, or use black-box deep learning features that lack interpretability. These shortcomings make it difficult for existing technologies to achieve accurate early disease identification and severity quantification, limiting their widespread application in actual agricultural production.

[0004] The information disclosed in the background section is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0005] The purpose of this invention is to provide a system and method for assessing the health status of tomato leaves, in order to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides a tomato leaf health status assessment system, specifically comprising:

[0007] The image processing module is used to perform size normalization, standardization and RGB to Lab color space conversion on the original image of the tomato leaf to be identified, generate a leaf region mask based on the a and b channel thresholds, and extract the first image containing only the tomato leaf region.

[0008] The pixel-level anomaly detection module is used to calculate the mean and standard deviation of the a and b channels in the first image, and set the sensitivity coefficient based on this to perform pixel-level anomaly detection. Based on the detection results, it generates anomaly area masks and normal area masks.

[0009] The region feature extraction module is used to perform connected component analysis on the abnormal region mask, extract each independent abnormal region, and extract its local color features and local texture features as recognition features, while extracting the corresponding features of the normal region as reference features.

[0010] The feature comparison and deviation calculation module is used to compare the identification features of each independent abnormal region with the reference features of the normal region item by item, calculate the feature distance set, and calculate the health deviation of each independent abnormal region.

[0011] The comprehensive health index generation module is used to normalize the health deviation of all independent abnormal areas and, in combination with the area ratio of abnormal areas of the tomato leaves to be identified, determine the comprehensive health assessment index used to characterize the health status of the leaves.

[0012] The health status judgment output module is used to judge the health status of the tomato leaves to be identified based on the comprehensive health assessment index and according to the preset judgment threshold rules, and output the overall health assessment result of the tomato leaves to be detected.

[0013] Furthermore, the specific process of the image processing module is as follows:

[0014] The original image of the tomato leaf to be identified is normalized by scaling it to a preset width W × height H pixel size using bilinear interpolation to eliminate the influence of image size differences. Then, normalization is performed by dividing the RGB three-channel pixel values ​​of the normalized image by 255 to convert them into floating-point numbers between 0 and 1. Finally, color space conversion is performed by using the CIE color space conversion standard to convert the normalized RGB image into a Lab color space image containing L, a, and b channels.

[0015] Then, based on the chroma components of the a and b channels of the Lab image, thresholds are set to generate a leaf region mask. The logic for generating the leaf region mask is as follows: set the thresholds for the a and b channels. When the a channel value of a pixel is less than the a channel threshold and the b channel value is less than the b channel threshold, the pixel is marked as a leaf region; otherwise, it is marked as a background region. The set of all pixels marked as leaf regions is set as the preliminary mask. The preliminary mask is then subjected to morphological processing of erosion followed by dilation to remove noise regions with an area smaller than the preset threshold and fill in the holes to obtain the final leaf region mask.

[0016] Preserve the Lab channel values ​​corresponding to the leaf region in the mask, set the Lab channel values ​​of the background region to 0, and perform a cropping operation to obtain a first image containing only the tomato leaf region.

[0017] Further, the a-channel and b-channel values ​​of all pixels in the first image are extracted, and the mean, standard deviation, mean, and standard deviation of the a-channel, a-channel, b-channel, and b-channel values ​​of all pixels are calculated.

[0018] Set a sensitivity coefficient k to control the anomaly detection range, where k > 0;

[0019] Iterate through each pixel in the first image and use the following rules to determine if the pixel has an abnormal color:

[0020] If the following condition is satisfied: |a p -μ a |>k*σ a or |b p -μ b |>k*σ b If so, the pixel is determined to be an abnormal pixel.

[0021] Otherwise, the pixel is determined to be a normal pixel;

[0022] Among them, a p μ is the a-channel value of pixel p, where p is the index of the pixel in the first image. a μ is the mean value of the a-channel of all pixels in the first image. a The standard deviation of channel a for all pixels in the first image; b p μ is the b-channel value of pixel p. b σ is the mean value of the b-channel of all pixels in the first image. b The standard deviation of the b-channel of all pixels in the first image;

[0023] Based on the detection results, abnormal region masks and normal region masks are generated. In the abnormal region mask, abnormal pixels are marked as 1 and normal pixels are marked as 0; in the normal region mask, normal pixels are marked as 1 and abnormal pixels are marked as 0.

[0024] Furthermore, a connected component analysis is performed on the abnormal region mask. The 8-neighborhood connectivity judgment rule is used to traverse all pixels in the abnormal region mask, and the interconnected abnormal pixels are divided into the same independent abnormal region. A unique identifier is assigned to each independent abnormal region.

[0025] Traverse all independent anomalous regions and extract local color features and local texture features for each independent anomalous region: the local color features include the mean and standard deviation of the L channel values ​​of all independent anomalous pixels in the Lab color space, and the local texture features include binary statistical data of the pixel color channels.

[0026] Local texture features are extracted as follows: For the three channels L, a, and b in the region, the difference between the pixel value and the mean value of the 8 neighboring pixels is calculated. When the absolute value of the difference is greater than 0, it is marked as 1; otherwise, it is marked as 0. The proportion of the number of pixels marked as 1 in each channel to the total number of pixels in the region is counted to obtain the binary statistical proportion values ​​corresponding to the three channels L, a, and b, which together constitute the local texture features.

[0027] The aforementioned local color features and local texture features are combined as the identification features of this independent anomaly region;

[0028] The mean and standard deviation of the L channel values ​​of each normal pixel in the entire normal region mask, as well as the binary statistical data of the pixel color channels, are extracted as reference features.

[0029] Furthermore, the specific execution process of the feature comparison and deviation calculation module is as follows:

[0030] First, it is clear that there are 5 characteristic items for independent abnormal regions to perform comparative analysis with normal regions, namely: the mean of L channel value, the standard deviation of L channel value, the binary statistical ratio of L channel, the binary statistical ratio of a channel, and the binary statistical ratio of b channel;

[0031] For each independent abnormal region, calculate the distance between each of its feature items and the corresponding feature items of the normal region; dynamically assign weights based on the standard deviation of each feature item in the normal region, so that the weights are inversely proportional to the standard deviation of the feature items and the sum of the weights is 1; sum the feature distances of each independent abnormal region according to the weights to obtain the health deviation of the region.

[0032] The formula used to calculate the distance between each feature term in an independent abnormal region and the corresponding feature term in a normal region is as follows:

[0033]

[0034] In the formula, d ijf represents the distance between the j-th feature term of the i-th independent anomalous region and the corresponding feature term in the reference feature of the normal region, used to quantify the degree of difference of a single feature term; ij f represents the value of the j-th feature term in the i-th independent anomaly region; oj ε represents the value of the j-th feature in the reference features of the normal region; ε is a very small positive number used to avoid the denominator being zero; i is the index of the independent abnormal region, j is the index of the feature, j = 1, 2, 3, 4, 5, which correspond to the mean value of the L channel, the standard deviation value of the L channel, the binary statistical proportion value of the L channel, the binary statistical proportion value of the a channel, and the binary statistical proportion value of the b channel, respectively;

[0035] The calculation of feature item weights is based on a pre-established feature library containing data from multiple healthy tomato leaf samples. In this feature library, the five feature items are extracted from each healthy leaf sample according to its normal region, and the standard deviation of each feature item in all healthy samples is calculated.

[0036] The formula for calculating the feature term weights is as follows:

[0037]

[0038] In the formula, W j σ represents the weight of the j-th feature, reflecting its importance in the health status assessment, and the sum of the weights of all features is 1; j is the standard deviation of the j-th feature term in the reference features of the normal region, used to characterize the degree of fluctuation of this feature under normal conditions;

[0039] The formula for calculating health deviation is as follows:

[0040]

[0041] In the formula, S i This represents the health deviation degree of the i-th independent abnormal region, used to comprehensively reflect the overall difference between this independent abnormal region and the normal region.

[0042] Furthermore, the health deviation of all independent abnormal regions is normalized by using preset global maximum health deviation value and global minimum health deviation value to convert the health deviation of each independent abnormal region to the [0,1] interval, and the mean of the normalized health deviation of all independent abnormal regions is calculated as the mean of the health deviation of the tomato leaf to be identified.

[0043] Obtain the area ratio of abnormal regions in the tomato leaf to be identified, whereby the area ratio of abnormal regions is the ratio of the total number of pixels in all independent abnormal regions to the total number of pixels in the leaf region.

[0044] Based on the mean value of health deviation and the proportion of the abnormal area, a comprehensive health assessment index for characterizing the health status of tomato leaves is determined, and the formula is as follows:

[0045]

[0046] In the formula, HEA is the comprehensive health assessment index, is the mean value of health deviation, R is the proportion of the abnormal area, α and β are preset weights, α + β = 1, and α > β > 0.

[0047] Furthermore, the comprehensive health assessment index is compared with a preset determination threshold to discriminate the health status of the to-be-identified tomato leaves. The specific logic is as follows:

[0048] If HEA ≤ H1, it is determined that the currently to-be-identified tomato leaf is in a severely abnormal state;

[0049] If H1 < HEA ≤ H2, it is determined that the currently to-be-identified tomato leaf is in a moderately abnormal state;

[0050] If H2 < HEA ≤ H3, it is determined that the currently to-be-identified tomato leaf is in a mildly abnormal state;

[0051] If HEA > H3, it is determined that the currently to-be-identified tomato leaf is in a healthy state;

[0052] In the formula, HEA is the comprehensive health assessment index, H1, H2, and H3 are preset determination thresholds, and H1 < H2 < H3.

[0053] The present invention further provides a method for evaluating the health status of tomato leaves. The method for evaluating the health status of tomato leaves is obtained by executing the above-mentioned system for evaluating the health status of tomato leaves. The specific steps include:

[0054] Step 1: Perform size normalization, standardization processing, and conversion from the RGB color space to the Lab color space on the original image of the to-be-identified tomato leaf, generate a leaf area mask based on the a and b channel thresholds, and extract the first image containing only the tomato leaf area;

[0055] Step 2: Statistically calculate the mean values and standard deviations of the a channel and the b channel in the first image, set a sensitivity coefficient based on this, perform pixel-level abnormal detection, and generate an abnormal area mask and a normal area mask according to the detection results;

[0056] Step 3: Perform connected component analysis on the abnormal area mask, extract each independent abnormal area, extract its local color features and local texture features as recognition features, and at the same time extract the corresponding features of the normal area as reference features;

[0057] Step 4: Compare the identification features of each independent abnormal region with the reference features of the normal region item by item, calculate the feature distance set, and calculate the health deviation of each independent abnormal region;

[0058] Step 5: Normalize the health deviation of all independent abnormal areas, and combine it with the area ratio of abnormal areas of the tomato leaves to be identified to determine the comprehensive health assessment index used to characterize the health status of the leaves.

[0059] Step 6: Based on the comprehensive health assessment index, and according to the preset judgment threshold rules, judge the health status of the tomato leaves to be identified, and output the overall health assessment result of the tomato leaves to be tested.

[0060] Compared with the prior art, the beneficial effects of the present invention are:

[0061] This invention utilizes dynamic threshold segmentation based on Lab color space and pixel-level anomaly detection to effectively eliminate background interference and accurately identify localized, minute anomaly regions on leaves, demonstrating extremely high sensitivity to early-stage diseases and overcoming the insensitivity of traditional methods to minor lesions. Secondly, by employing a region comparison and weighted deviation calculation mechanism based on statistical features, it can accurately quantify the severity of each anomaly region without relying on complex deep learning models and large amounts of labeled data, significantly reducing system computational complexity and deployment costs while ensuring the objectivity and interpretability of the assessment results. Finally, by fusing the mean health deviation and the proportion of anomaly area to generate a comprehensive health index, and using this index for multi-threshold state discrimination, the assessment results not only reflect the severity of the disease but also its spatial distribution, providing agricultural producers with comprehensive and accurate decision-making support and greatly improving the automation level and assessment accuracy of tomato disease monitoring. Attached Figure Description

[0062] Figure 1 This is a schematic diagram of the overall system modules of the present invention;

[0063] Figure 2 This is a schematic diagram of the overall method flow of the present invention;

[0064] Figure 3 The graph is a function of the mean health deviation, the percentage of abnormal area, and the comprehensive health assessment index.

[0065] Figure 4 This is a parallel coordinate graph of the mean health deviation, the percentage of abnormal area, and the comprehensive health assessment index. Detailed Implementation

[0066] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0067] It should be noted that, unless otherwise defined, the technical or scientific terms used in this invention should have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed following the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0068] Example:

[0069] Please see Figure 1 This invention provides a tomato leaf health status assessment system, specifically comprising:

[0070] The image processing module is used to perform size normalization, standardization and RGB to Lab color space conversion on the original image of the tomato leaf to be identified, generate a leaf region mask based on the a and b channel thresholds, and extract the first image containing only the tomato leaf region.

[0071] In this embodiment, the specific process of the image processing module is as follows:

[0072] The original image of the tomato leaf to be identified is normalized by scaling it to a preset width W × height H pixel size using bilinear interpolation to eliminate the influence of image size differences. Then, normalization is performed by dividing the RGB three-channel pixel values ​​of the normalized image by 255 to convert them into floating-point numbers between 0 and 1. Finally, color space conversion is performed by using the CIE color space conversion standard to convert the normalized RGB image into a Lab color space image containing L, a, and b channels.

[0073] Then, based on the chroma components of the a and b channels of the Lab image, thresholds are set to generate a leaf region mask. The logic for generating the leaf region mask is as follows: set the a-channel threshold and the b-channel threshold. When the a-channel value of a pixel is less than the a-channel threshold and the b-channel value is less than the b-channel threshold, the pixel is marked as a leaf region; otherwise, it is marked as a background region. The set of all pixels marked as leaf regions is set as the initial mask. The initial mask is first subjected to an erosion operation, using a structuring element of a specific size to traverse the image, eliminate isolated noise points and shrink the region boundaries. Then, a dilation operation is performed, using the same structuring element to traverse the image, restore the original size of the effective leaf region and fill the holes inside the region. Finally, area filtering is used to remove tiny regions with an area smaller than the preset number of pixels that still exist after morphological processing, thereby obtaining a complete and accurate final leaf region mask.

[0074] The thresholds for channel a and channel b are determined based on the chromaticity statistical characteristics of healthy tomato leaf samples: the threshold for channel a is negative, typically set in the range of -5 to -10, to capture the typical green characteristics of the leaves; the threshold for channel b is a lower positive value, typically set in the range of 10 to 20, to exclude common high-yellow background interference. Specific thresholds are determined by statistically analyzing the specific quantiles of the channel a and b values ​​of healthy leaf samples and fine-tuning them in conjunction with the actual application scenario.

[0075] Preserve the Lab channel values ​​corresponding to the leaf region in the mask, set the Lab channel values ​​of the background region to 0, and perform a cropping operation to obtain a first image containing only the tomato leaf region.

[0076] In this embodiment, the main reason for converting the RGB image to the Lab color space is its superior color perception uniformity and excellent color-luminance separation characteristics. The Lab color space completely separates luminance information (L channel) from color information (a and b channels). The a and b channels can more accurately represent the green features of the leaves, making it more accurate and stable to distinguish green leaves from non-green backgrounds by setting thresholds based on chromaticity components. This effectively reduces the impact of illumination changes on color features, laying a reliable foundation for subsequent accurate segmentation and anomaly detection.

[0077] The image processing module of this system employs a strategy combining dual-threshold interval segmentation (a and b channels) in the Lab color space with morphological post-processing. Its core advantage lies in its precise extraction capability of leaf regions. Compared to existing methods that commonly use single-threshold segmentation or direct processing in the RGB space, this module effectively locks green vegetation regions using the negative values ​​of the a channel, while utilizing the positive values ​​of the b channel to eliminate yellow background interference such as soil and flowerpots. Morphological operations are then used to remove noise and fill holes, and finally, masking operations are used to precisely separate the pure leaf regions. This processing method significantly improves the accuracy and reliability of subsequent analysis.

[0078] Compared with existing technologies, the benefits of this module are mainly reflected in three aspects: First, it solves the problem of incomplete leaf extraction in complex backgrounds and improves the accuracy of leaf region segmentation; second, the use of the Lab color space greatly reduces the sensitivity of color features to changes in illumination and improves the stability of the system under different illumination conditions; third, through accurate leaf region extraction, it effectively avoids the interference of background noise on subsequent anomaly detection and provides a high-quality data foundation for pixel-level anomaly detection.

[0079] As the front-end preprocessing stage of the entire system, the quality of this module directly determines the accuracy of all subsequent analyses. Precise leaf region extraction provides clean input data for subsequent anomaly detection, ensuring the reliability of statistical features; high-quality image preprocessing reduces the generation of false anomalies, improving the system's sensitivity to detecting real diseases; simultaneously, stable color space conversion provides consistent color representation for subsequent feature extraction, ensuring the system's repeatability and stability under different acquisition conditions, laying a solid foundation for the accurate operation of the entire evaluation system.

[0080] The pixel-level anomaly detection module is used to calculate the mean and standard deviation of the a and b channels in the first image, and set the sensitivity coefficient based on this to perform pixel-level anomaly detection. Based on the detection results, it generates anomaly area masks and normal area masks.

[0081] In this embodiment, the a-channel and b-channel values ​​of all pixels in the first image are extracted, and the mean, standard deviation, mean, and standard deviation of the a-channel, a-channel, b-channel, and b-channel values ​​of all pixels are calculated.

[0082] Set a sensitivity coefficient k to control the anomaly detection range, where k > 0;

[0083] Iterate through each pixel in the first image and use the following rules to determine if the pixel has an abnormal color:

[0084] If the following condition is satisfied: |a p -μ a |>k*σa or |b p -μ b |>k*σ b If so, the pixel is determined to be an abnormal pixel.

[0085] Otherwise, the pixel is determined to be a normal pixel;

[0086] Among them, a p μ is the a-channel value of pixel p, where p is the index of the pixel in the first image. a μ is the mean value of the a-channel of all pixels in the first image. a The standard deviation of channel a for all pixels in the first image; b p μ is the b-channel value of pixel p. b σ is the mean value of the b-channel of all pixels in the first image. b is the standard deviation of the b-channel of all pixels in the first image.

[0087] The reason for adopting this detection rule is that it is based on statistical principles and can achieve adaptive abnormal pixel identification. This method establishes a statistical distribution model of normal color by calculating the mean and standard deviation of the color channels in the leaf region, and dynamically sets the threshold range for anomaly detection using a sensitivity coefficient k. When the a or b channel value of a pixel deviates from the mean by more than this dynamic threshold, it is judged as abnormal. This allows the detection process to automatically adapt to the overall color differences in different images without setting a fixed threshold. It ensures high sensitivity to early, subtle color anomalies while effectively reducing false detections caused by inherent differences in lighting, variety, etc., significantly improving the accuracy and robustness of the detection.

[0088] Based on the detection results, abnormal region masks and normal region masks are generated. In the abnormal region mask, abnormal pixels are marked as 1 and normal pixels are marked as 0; in the normal region mask, normal pixels are marked as 1 and abnormal pixels are marked as 0.

[0089] The pixel-level anomaly detection module of this system employs a dynamic threshold detection method based on statistical process control. Its core advantage lies in achieving adaptive anomaly pixel recognition. This module calculates the mean and standard deviation of channels a and b of the leaf region, combined with an adjustable sensitivity coefficient k, to dynamically generate a detection threshold adapted to the color distribution characteristics of each image to be detected. This detection method, based on the statistical characteristics of the image itself, can effectively handle the natural color changes of tomato leaves of different varieties and at different growth stages, avoiding the limitations of fixed threshold methods.

[0090] Compared with existing technologies, the beneficial effects of this module are mainly reflected in three aspects: First, the use of dynamic thresholds instead of fixed thresholds greatly improves the system's adaptability to different imaging conditions and leaf physiological states; second, anomaly detection at the pixel level can detect the earliest minute lesions, enabling early warning of diseases; third, the adjustable design of the sensitivity coefficient k allows the system to flexibly balance detection sensitivity and false alarm rate according to actual application needs, improving the system's practicality and operability.

[0091] This module serves as a crucial link between the preceding and following steps, playing a vital bridging role in the overall solution. Precise pixel-level anomaly detection provides high-quality input data for subsequent region analysis, ensuring the accuracy of anomaly region segmentation; adaptive detection thresholds reduce reliance on manual parameter tuning, improving the system's automation level; and the ability to detect early, minute anomalies enables the system to promptly identify initial lesions, providing a window for early intervention, thereby enhancing the overall system's practical value and early warning capabilities.

[0092] The region feature extraction module is used to perform connected component analysis on the abnormal region mask, extract each independent abnormal region, and extract its local color features and local texture features as recognition features, while extracting the corresponding features of the normal region as reference features.

[0093] In this embodiment, connected component analysis is performed on the abnormal region mask. The 8-neighborhood connectivity judgment rule is used to traverse all pixels in the abnormal region mask, and the interconnected abnormal pixels are divided into the same independent abnormal region. A unique identifier is assigned to each independent abnormal region.

[0094] Traverse all independent anomalous regions and extract local color features and local texture features for each independent anomalous region: the local color features include the mean and standard deviation of the L channel values ​​of all independent anomalous pixels in the Lab color space, and the local texture features include binary statistical data of the pixel color channels.

[0095] Local texture features are extracted as follows: For the three channels L, a, and b in the region, the difference between the pixel value and the mean value of the 8 neighboring pixels is calculated. When the absolute value of the difference is greater than 0, it is marked as 1; otherwise, it is marked as 0. The proportion of the number of pixels marked as 1 in each channel to the total number of pixels in the region is counted to obtain the binary statistical proportion values ​​corresponding to the three channels L, a, and b, which together constitute the local texture features.

[0096] The aforementioned local color features and local texture features are combined as the identification features of this independent anomaly region;

[0097] The mean and standard deviation of the L channel values ​​of each normal pixel in the entire normal region mask, as well as the binary statistical data of the pixel color channels, are extracted as reference features.

[0098] The system's region feature extraction module employs a strategy of connected component analysis and multi-feature fusion, its core advantage being the precise conversion from pixel-level detection to region-level analysis. This module performs 8-neighborhood connectivity analysis on anomaly region masks, aggregating discrete anomalous pixels into meaningful independent anomalous regions, and extracting a composite feature vector for each region containing L-channel statistical features and three-channel texture features. This regionalized processing approach preserves local detail information while forming a complete region representation, laying the foundation for subsequent quantitative evaluation.

[0099] Compared with existing technologies, the beneficial effects of this module are mainly reflected in three aspects: First, by using connected component analysis, pixel detection results are transformed into physically meaningful region objects, avoiding the oversensitivity to isolated noise points in traditional methods; second, by simultaneously extracting color statistical features and texture structure features, abnormal region characteristics are described from multiple dimensions, providing richer discriminative information than single features; third, by quantifying texture features through binary statistical ratio values, both the interpretability of the features and the adaptability to abnormal regions at different scales are ensured.

[0100] This module plays a crucial role in the overall solution, serving as a bridge between previous and subsequent steps. By aggregating pixel-level detection results into region objects, it provides a clear analytical unit for subsequent feature comparison and health assessment. The extraction of multi-dimensional features provides a sufficient data foundation for health deviation calculation, ensuring the comprehensiveness and accuracy of the evaluation results. At the same time, the synchronous extraction of reference features for normal regions establishes a reliable comparison benchmark, giving subsequent deviation calculations a clear reference standard, thereby significantly improving the scientific rigor and reliability of the entire system's evaluation results.

[0101] The feature comparison and deviation calculation module is used to compare the identification features of each independent abnormal region with the reference features of the normal region item by item, calculate the feature distance set, and calculate the health deviation of each independent abnormal region.

[0102] In this embodiment, the specific execution process of the feature comparison and deviation calculation module is as follows:

[0103] First, it is clear that there are 5 characteristic items for independent abnormal regions to perform comparative analysis with normal regions, namely: the mean of L channel value, the standard deviation of L channel value, the binary statistical ratio of L channel, the binary statistical ratio of a channel, and the binary statistical ratio of b channel;

[0104] For each independent abnormal region, calculate the distance between each of its feature items and the corresponding feature items of the normal region; dynamically assign weights based on the standard deviation of each feature item in the normal region, so that the weights are inversely proportional to the standard deviation of the feature items and the sum of the weights is 1; sum the feature distances of each independent abnormal region according to the weights to obtain the health deviation of the region.

[0105] The formula used to calculate the distance between each feature term in an independent abnormal region and the corresponding feature term in a normal region is as follows:

[0106]

[0107] In the formula, d ij f represents the distance between the j-th feature term of the i-th independent anomalous region and the corresponding feature term in the reference feature of the normal region, used to quantify the degree of difference of a single feature term; ij f represents the value of the j-th feature term in the i-th independent anomaly region; oj ε represents the value of the j-th feature in the reference features of the normal region; ε is a very small positive number used to avoid the denominator being zero; i is the index of the independent abnormal region, and j is the index of the feature, j = 1, 2, 3, 4, 5, which correspond to the mean value of the L channel, the standard deviation value of the L channel, the binary statistical proportion value of the L channel, the binary statistical proportion value of the a channel, and the binary statistical proportion value of the b channel, respectively.

[0108] The dependent variable d in the formula ij This represents the relative distance; a larger value indicates a more significant difference between the i-th anomalous region and the normal reference feature in the j-th feature. This value quantifies the degree of deviation of a specific feature of the anomalous region, and its value is positively correlated with the severity of the anomalousness. The formula is reasonable: using absolute values ​​ensures the non-negativity of the distance; and using the denominator f... oj +ε enables relative distance calculation, which eliminates the influence of different feature dimensions and avoids division-to-zero errors through the minimal constant ε. The resulting dimensionless result facilitates comparison between different feature terms and subsequent weighted fusion, providing standardized input for the calculation of health deviation.

[0109] The calculation of feature item weights is based on a pre-established feature library containing data from multiple healthy tomato leaf samples. In this feature library, the five feature items are extracted from each healthy leaf sample according to its normal region, and the standard deviation of each feature item in all healthy samples is calculated.

[0110] The formula for calculating the feature term weights is as follows:

[0111]

[0112] In the formula, W jσ represents the weight of the j-th feature, reflecting its importance in the health status assessment, and the sum of the weights of all features is 1; j is the standard deviation of the j-th feature term in the reference features of the normal region, used to characterize the degree of fluctuation of this feature under normal conditions.

[0113] The formula for calculating the weight of this feature term is based on the principle of statistical reliability. Its core meaning is that the weight of a feature term is inversely proportional to the square of its fluctuation under healthy conditions. Standard deviation σ j The smaller the value, the more stable and consistent the feature is in healthy samples, and the stronger its reliability in identifying abnormal states, thus it is assigned a higher weight; conversely, the greater the fluctuation, the lower the weight of the feature. The formula ensures that the sum of all weights is 1 through normalization, realizing the automatic allocation of importance based on the objective differences in the identification capabilities of different features, making health assessment more scientific and accurate.

[0114] The formula for calculating health deviation is as follows:

[0115]

[0116] In the formula, S i This represents the health deviation degree of the i-th independent abnormal region, used to comprehensively reflect the overall difference between this independent abnormal region and the normal region.

[0117] The dependent variable S in the formula i The weighted average distance represents the distance between the i-th abnormal region and the normal region across multiple feature dimensions. A larger value indicates a more significant overall difference in health status between the i-th abnormal region and the normal region across multiple feature dimensions, meaning a more severe deviation in the health status of that region. This formula is reasonable: it uses a weighted summation method to integrate deviation information from five feature dimensions, considering both the differences in importance of different feature items and preserving the contribution of each dimension's deviation degree. The result is a comprehensive index that quantifies the degree of abnormality, providing a reliable and interpretable numerical basis for subsequent health status assessments.

[0118] The feature comparison and deviation calculation module of this system employs a statistically weighted multi-dimensional feature distance quantification method, whose core advantage lies in achieving a scientific and objective quantitative assessment of health status. This module establishes a five-dimensional feature comparison system, combined with a reverse weighting mechanism based on the standard deviation of a health sample database. This approach considers both the differences in the discriminative power of different feature items and ensures the objectivity of weight allocation. The use of the relative distance calculation formula effectively eliminates the influence of feature dimensions, while the introduction of a minimal constant ε ensures the numerical stability of the calculation.

[0119] Compared with existing technologies, the beneficial effects of this module are mainly reflected in three aspects: First, the use of multi-dimensional feature weighted fusion to replace single feature judgment significantly improves the accuracy and robustness of health status assessment; second, the weight allocation method based on the statistical characteristics of the health sample library overcomes the subjectivity of manually setting weights, making the assessment results more scientific and reliable; third, through the quantitative calculation of health deviation, the transformation from qualitative judgment to quantitative assessment is realized, providing accurate data support for the classification of disease severity.

[0120] This module serves as the core computing engine in the overall solution. The multi-feature weighted fusion mechanism fully utilizes the various feature information extracted in the early stages, maximizing the discriminative value of the feature data. An objective weight allocation method ensures the consistency of evaluation results across different blades and under different conditions, improving the system's reliability. The accurate calculation of health deviation provides crucial input parameters for the subsequent generation of the comprehensive health index, ensuring that the final health status judgment is based on rigorous quantitative principles, thereby guaranteeing the scientific rigor and accuracy of the entire system's evaluation results.

[0121] The comprehensive health index generation module is used to normalize the health deviation of all independent abnormal areas and, in combination with the area ratio of abnormal areas of the tomato leaves to be identified, determine the comprehensive health assessment index used to characterize the health status of the leaves.

[0122] In this embodiment, the health deviation of all independent abnormal regions is normalized. The preset global maximum health deviation value and global minimum health deviation value are used to convert the health deviation of each independent abnormal region to the [0,1] interval, and the mean of the normalized health deviation of all independent abnormal regions is calculated as the mean of the health deviation of the tomato leaf to be identified.

[0123] Obtain the area ratio of abnormal regions in the tomato leaf to be identified, whereby the area ratio of abnormal regions is the ratio of the total number of pixels in all independent abnormal regions to the total number of pixels in the leaf region.

[0124] Based on the mean health deviation and the proportion of abnormal area, a comprehensive health assessment index for characterizing the health status of tomato leaves is determined, using the following formula:

[0125]

[0126] In the formula, HEA is the comprehensive health assessment index. Let R be the mean of the health deviation, R be the percentage of the abnormal area, and α and β be preset weights, where α + β = 1 and α > β > 0. The reason for setting α > β is that the mean of the health deviation directly reflects the severity of the abnormal area and the degree of physiological damage, serving as a core qualitative indicator for assessing health status; while the percentage of the abnormal area mainly reflects the spatial distribution of the disease, serving as a quantitative auxiliary indicator. In agricultural pathology practice, a small but highly severe lesion often has a far greater destructive effect on leaf photosynthesis and overall physiological function than a large but slightly discolored area. Therefore, α is assigned a weight that reflects the severity of the abnormality. Higher weighting can more accurately represent the true health level of leaves, which is in line with the key considerations of plant protection experts when making decisions.

[0127] The dependent variable HEA in the formula is the Comprehensive Health Assessment Index, which specifically represents the overall health status of tomato leaves. This index uses mathematical transformation to integrate two key indicators—the mean of health deviation and the proportion of abnormal area—into a comprehensive evaluation value. Its technical advantage lies in providing a standardized assessment indicator with a defined range and strong comparability. A higher value indicates healthier leaves, while a lower value indicates more severe disease, thus achieving a precise quantitative assessment of health status.

[0128] Independent variable The HEA index reflects the severity of the abnormal area, while R reflects the extent of disease spread. Both describe leaf health from qualitative and quantitative dimensions, respectively. The HEA index is intrinsically related to these two independent variables because leaf health depends on both the severity of the abnormal area and the proportion of the abnormal part, which together determine the overall physiological function and health of the leaf.

[0129] HEA Index and Independent Variables Both R and R show a negative correlation: when When R increases, the formula... When the term decreases, the HEA index decreases accordingly after cube root and logarithmic transformations, indicating a deterioration in health status; conversely, when the term increases... When R decreases, the HEA index increases, indicating improved health status. This negative correlation aligns with the principles of plant pathology and accurately reflects the intrinsic link between disease severity and health status.

[0130] The formula has multiple rationale factors: First, it uses a reciprocal form to convert the mean health deviation and the percentage of abnormal area, two indicators negatively correlated with health status, into positively correlated variables, which aligns with the logic that a higher health index indicates better health; second, it uses a weighted summation... The system integrates information from two dimensions, and the weighting reflects the difference in importance between the two. The cube root operation compresses the data range to avoid the influence of extreme values, while the logarithmic transformation further standardizes the output distribution, making the HEA value range smoother and more stable. Finally, adding 1 and taking the logarithm ensures that the result is always positive. The entire transformation process maintains the monotonicity of the function, and finally outputs a standardized and meaningful comprehensive health assessment index.

[0131] Table 1: Comprehensive Health Assessment Index Statistics Table

[0132] Data group number Mean of health deviation percentage of abnormal area Comprehensive Health Assessment Index 1 0.05 0.01 2.43 2 0.1 0.03 1.93 3 0.15 0.05 1.63 4 0.2 0.07 1.41 5 0.25 0.1 1.23 6 0.3 0.15 1.05 7 0.4 0.2 0.85 8 0.5 0.25 0.7 9 0.6 0.3 0.58 10 0.7 0.35 0.48 11 0.75 0.4 0.42 12 0.8 0.5 0.34 13 0.85 0.6 0.27 14 0.9 0.7 0.21 15 0.95 0.8 0.15

[0133] Please refer to the following: Figures 3-4 The comprehensive health assessment index showed a significant negative correlation with the mean health deviation and the proportion of abnormal area. Data analysis indicates that when the leaf health status is good, i.e. Both the R value and the HEA index are at low levels, while the HEA index remains in a high range, reflecting that the overall physiological function of the leaves is normal; as the severity of the disease increases, As the R value gradually increases, the HEA index shows a smooth and continuous downward trend, demonstrating the indicator's good responsiveness to changes in health status.

[0134] The changing pattern of the HEA index fully reflects the rationality of the formula design: its value range is relatively wide, which can clearly distinguish different state levels from healthy to severely abnormal; the index decline curve shows a characteristic of being rapid at first and then slow, indicating that the index has high sensitivity in the early stage of disease and can effectively identify early abnormalities, while it can still maintain a certain degree of differentiation when the disease is more severe, avoiding numerical saturation, and providing a reliable basis for accurate judgment of health status.

[0135] The comprehensive health index generation module of this system adopts a multi-parameter fusion nonlinear modeling method, whose core advantage lies in realizing a comprehensive and quantitative assessment of the health status of leaves. This module organically combines two key indicators—the mean health deviation and the proportion of abnormal area—and uses a special logarithmic cube root function to reflect both the severity of the abnormal area and the spread of the abnormal range, ultimately generating a comprehensive health assessment index with clear physical meaning.

[0136] Compared with the prior art, the beneficial effects brought by this module are mainly reflected in three aspects: First, by considering both the mean value of health deviation and the proportion of abnormal area, it overcomes the one-sidedness of single-index evaluation and provides a more comprehensive description of the health status; Second, through normalization using preset global maximum and minimum values, it ensures the comparability and consistency of evaluation results among different leaves; Third, the special function form design not only ensures the smoothness of the index but also enhances the sensitivity to distinguish severe disease conditions, making the evaluation results more in line with the actual needs of agricultural production.

[0137] This module plays a key role in information fusion and decision preparation in the overall scheme. By integrating the health deviation degrees of multiple abnormal regions, it condenses the complex regional-level evaluation results into a single comprehensive index, greatly simplifying the subsequent discrimination and decision-making process; The scientifically designed index calculation formula ensures that the evaluation results are neither overly sensitive nor overly insensitive, and can provide reasonable evaluation values in the case of diseases of different severities; The finally generated HEA index provides an intuitive and reliable input basis for the health status discrimination module, making the output results of the entire system more concise and clear, facilitating practical application and promotion.

[0138] The health status discrimination output module is used to discriminate the health status of the to-be-identified tomato leaf according to the preset determination threshold rule based on the comprehensive health evaluation index, and output the overall health evaluation result of the to-be-detected tomato leaf;

[0139] In this embodiment, the comprehensive health evaluation index is compared with the preset determination threshold to discriminate the health status of the to-be-identified tomato leaf. The specific logic is as follows:

[0140] If HEA ≤ H1, it is determined that the currently to-be-identified tomato leaf is in a severely abnormal state;

[0141] If H1 < HEA ≤ H2, it is determined that the currently to-be-identified tomato leaf is in a moderately abnormal state;

[0142] If H2 < HEA ≤ H3, it is determined that the currently to-be-identified tomato leaf is in a mildly abnormal state;

[0143] If HEA > H3, it is determined that the currently to-be-identified tomato leaf is in a healthy state;

[0144] In the formula, HEA is the comprehensive health evaluation index, H1, H2, and H3 are preset determination thresholds, and H1 < H2 < H3.

[0145] The determination of H1, H2, and H3 is based on the statistical results of a large number of tomato leaf samples with known health status: by calculating the comprehensive health assessment index of these samples, the lower limit of the HEA value of healthy samples is selected as H3, the lower limit of the HEA value of slightly abnormal samples is selected as H2, and the lower limit of the HEA value of moderately abnormal samples is selected as H1, so as to ensure that each threshold is located at the typical numerical boundary point of different health status levels. Finally, the initial thresholds obtained from the statistics are fine-tuned and determined by cluster analysis and expert experience method.

[0146] The health status assessment output module of this system employs a multi-threshold hierarchical assessment strategy. Its core advantage lies in achieving refined classification and intuitive output of health status. By setting three scientifically graded assessment thresholds, this module maps the continuous Comprehensive Health Assessment Index (HEA) to four distinct health levels, ensuring both accuracy and clear classification results. The module output not only includes the final health status assessment but also provides detailed evaluation parameters, forming a complete assessment report system.

[0147] Compared with existing technologies, the benefits of this module are mainly reflected in three aspects: First, by using multi-level threshold judgment to replace the traditional binary classification method, it can more precisely distinguish different degrees of abnormal health states, providing a basis for hierarchical management; second, the standardized setting of preset thresholds ensures the consistency and comparability between different assessment results, avoiding the bias of subjective judgment; third, the comprehensive output of health status level and detailed parameter indicators not only meets the need for rapid judgment, but also provides data support for in-depth analysis, greatly improving the practicality of the system.

[0148] This module plays a crucial role in realizing value and delivering results within the overall solution. By transforming quantified health indices into intuitive health status levels, it makes complex calculations and analyses easier to understand and use, significantly improving system usability. Standardized judgment rules ensure the objectivity and repeatability of assessment results, providing a unified evaluation standard for results under different times and conditions. The complete output system not only serves immediate decision-making but also provides a data foundation for long-term health monitoring and trend analysis, ultimately achieving a closed-loop value chain from image processing to decision support.

[0149] Please see Figure 2 A method for assessing the health status of tomato leaves, the specific steps of which include:

[0150] Step 1: Perform size normalization, standardization, and RGB to Lab color space conversion on the original image of the tomato leaf to be identified. Generate a leaf region mask based on the thresholds of the a and b channels, and extract the first image containing only the tomato leaf region.

[0151] Step 2: Calculate the mean and standard deviation of channels a and b in the first image, and set the sensitivity coefficient based on this to perform pixel-level anomaly detection. Generate anomaly area masks and normal area masks based on the detection results.

[0152] Step 3: Perform connected component analysis on the abnormal region mask, extract each independent abnormal region, and extract its local color features and local texture features as recognition features. At the same time, extract the corresponding features of the normal region as reference features.

[0153] Step 4: Compare the identification features of each independent abnormal region with the reference features of the normal region item by item, calculate the feature distance set, and calculate the health deviation of each independent abnormal region;

[0154] Step 5: Normalize the health deviation of all independent abnormal areas, and combine it with the area ratio of abnormal areas of the tomato leaves to be identified to determine the comprehensive health assessment index used to characterize the health status of the leaves.

[0155] Step 6: Based on the comprehensive health assessment index, and according to the preset judgment threshold rules, judge the health status of the tomato leaves to be identified, and output the overall health assessment result of the tomato leaves to be tested.

[0156] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0157] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented in software, the above embodiments can be implemented, in whole or in part, as a computer program product. Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution.

[0158] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0159] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A system for assessing the health status of tomato foliage, characterized by, Specifically comprising: An image processing module, configured to perform size normalization, standardization processing and RGB to Lab color space conversion on an original image of a tomato leaf to be identified, generate a leaf region mask based on a and b channel threshold values, and extract a first image containing only a tomato leaf region; A pixel-level anomaly detection module, configured to count the mean and standard deviation of the a channel and the b channel in the first image, set a sensitivity coefficient based on the mean and the standard deviation, perform pixel-level anomaly detection, and generate an abnormal region mask and a normal region mask according to a detection result; A region feature extraction module, configured to perform connected domain analysis on the abnormal region mask, extract each independent abnormal region, and extract local color features and local texture features of the independent abnormal region as identification features, and extract corresponding features of the normal region as reference features; A feature comparison and deviation degree calculation module, configured to compare the identification features of each independent abnormal region with the reference features of the normal region item by item, calculate a feature distance set, and calculate a health deviation degree of each independent abnormal region, and the specific execution process comprises: First, it is clear that the independent abnormal region has five feature items for comparison and analysis with the normal region, which are: the mean of the L channel value, the standard deviation of the L channel value, the L channel binary statistical proportion value, the a channel binary statistical proportion value, and the b channel binary statistical proportion value; For each independent abnormal region, the distance between each feature item and the corresponding feature item of the normal region is calculated; the standard deviation of each feature item of the normal region is dynamically assigned a weight, so that the weight is inversely proportional to the feature item standard deviation and the sum of the weights is 1; the feature distance of each independent abnormal region is weighted and summed according to the weight to obtain the health deviation degree of the region; A comprehensive health index generation module, configured to normalize the health deviation degrees of all independent abnormal regions, and determine a comprehensive health evaluation index for representing the health status of the leaf based on the area proportion of the abnormal region of the tomato leaf to be identified, and the specific execution process comprises: Calculate the mean of the normalized health deviation degrees of all independent abnormal regions as the mean of the health deviation degrees of the tomato leaf to be identified; According to the mean of the health deviation degrees and the area proportion of the abnormal region, a comprehensive health evaluation index for representing the health status of the tomato leaf is determined, and the formula is as follows: In the formula, is a comprehensive health assessment index, is a health deviation degree average, is an abnormal area area proportion, and is a preset weight, , and ; A health status discrimination output module, configured to discriminate the health status of the tomato leaf to be identified according to the comprehensive health evaluation index and a preset discrimination threshold rule, and output an overall health evaluation result of the tomato leaf to be detected.

2. The system for evaluating the health status of tomato leaves according to claim 1, wherein: The specific process of the image processing module is as follows: The original image of the to-be-identified tomato leaf is subjected to size normalization processing, and is scaled to a preset width through a bilinear interpolation method × height Pixel size, to eliminate the influence of image size difference; then, standardization processing is performed, the RGB three-channel pixel values of the size normalized image are divided by 255 respectively, and converted into floating-point numbers between 0 and 1; subsequently, color space conversion is performed, and the standardized RGB image is converted into a Lab color space image containing L, a and b channels through a CIE color space conversion standard; Then, based on the chroma components of the a channel and the b channel of the Lab image, threshold values are set to generate a leaf region mask, and the logic for generating the leaf region mask is as follows: set the a channel threshold value and the b channel threshold value, when the a channel value of a pixel point is less than the a channel threshold value and the b channel value is less than the b channel threshold value, mark the pixel point as a leaf region, otherwise mark it as a background region, and set the set of all pixel points marked as a leaf region as a preliminary mask; Perform morphological processing of the preliminary mask by first eroding and then dilating to remove noise regions with an area less than a preset threshold and fill holes to obtain a final leaf region mask; Reserve the Lab channel value corresponding to the leaf area in the mask, set the Lab channel value of the background area to 0, and perform a cutting operation to obtain a first image containing only the tomato leaf area.

3. A system for assessing the health status of tomato foliage according to claim 2, characterized in that: Extract the a channel and b channel values of all pixel points in the first image, and calculate the a channel mean, a channel standard deviation, b channel mean, and b channel standard deviation of all pixel points. Setting a sensitivity coefficient for controlling an abnormality detection range , and ; Traverse each pixel point in the first image, and use the following rules to determine whether the pixel point is color abnormal: If the following condition is satisfied: or then the pixel point is determined as an abnormal pixel point. Otherwise, determine that the pixel point is a normal pixel point. wherein, is the a channel value of the pixel point p, p is the index of the pixel point in the first image, is the a channel mean value of all pixel points in the first image, is the a channel standard deviation of all pixel points in the first image; is the b channel value of the pixel point p, is the b channel mean value of all pixel points in the first image, is the b channel standard deviation of all pixel points in the first image. Generate an abnormal area mask and a normal area mask according to the detection result, wherein in the abnormal area mask, the abnormal pixel points are marked as 1, and the normal pixel points are marked as 0; in the normal area mask, the normal pixel points are marked as 1, and the abnormal pixel points are marked as 0.

4. The system for assessing the health status of tomato foliage according to claim 3, wherein: Perform connected component analysis on the abnormal area mask, traverse all pixel points in the abnormal area mask using an 8-neighbor connectedness judgment rule, divide the mutually connected abnormal pixel points into the same independent abnormal area, and assign a unique identifier to each independent abnormal area. Traverse all independent abnormal areas, and extract local color features and local texture features for each independent abnormal area: the local color features include the mean and standard deviation of the L channel value of all independent abnormal pixels in the region in the Lab color space, and the local texture features include the binary statistical data of the pixel color channel. The local texture features are extracted by the following method: for the L, a, and b channels in the region, calculate the difference between the pixel value and the average value of the 8-neighbor pixels, mark 1 when the absolute value of the difference is greater than 0, otherwise mark 0, and calculate the proportion of the number of pixel points marked as 1 in the total number of pixels in each channel to obtain the binary statistical proportion value of the L, a, and b channels, which together constitute the local texture features. Combine the above local color features and local texture features as the identification features of the independent abnormal area. Extract the mean and standard deviation of the L channel value of each normal pixel point in the entire normal area mask in the Lab color space, and the binary statistical data of the pixel color channel as reference features.

5. The system for assessing the health status of tomato foliage according to claim 1, wherein: The specific execution process of the feature comparison and deviation degree calculation module is as follows: First, it is clear that there are five feature items for independent abnormal areas to compare with the normal area for analysis, which are: the mean of the L channel value, the standard deviation of the L channel value, the L channel binary statistical proportion value, the a channel binary statistical proportion value, and the b channel binary statistical proportion value. For each independent abnormal area, calculate the distance between each feature item and the corresponding feature item of the normal area. Based on the standard deviation of each feature item of the normal area, dynamically assign weights, so that the weight is inversely proportional to the feature item standard deviation and the sum of the weights is 1; weight the feature distance of each independent abnormal area according to the weight to obtain the health deviation degree of the region. The formula for calculating the distance between each feature item of the independent abnormal area and the corresponding feature item of the normal area is as follows: In the formula, represents the distance between the jth feature item of the ith independent abnormal region and the corresponding feature item in the normal region reference feature, which is used to quantify the difference degree of a single feature item; is the value of the jth feature item of the ith independent abnormal region; is the value of the jth feature item in the normal region reference feature; is a very small positive number, used to avoid the denominator being zero; i is the index of the independent abnormal region, is the index of the feature item, respectively correspond to the mean value of the L channel value, the standard deviation of the L channel value, the L channel binary statistical proportion value, the a channel binary statistical proportion value, and the b channel binary statistical proportion value. The calculation of the feature item weight is based on a pre-established feature library containing multiple healthy tomato leaf sample data, in which the five feature items are extracted from the normal area of each healthy leaf sample, and the standard deviation of the value of each feature item in all healthy tomato leaf samples is calculated; The calculation formula of the feature item weight is as follows: In the formula, represents the weight of the jth feature item, used to reflect the importance of the feature item in the health status evaluation, and the sum of all feature item weights is 1; is the standard deviation of the jth feature item in the normal region reference feature, used to represent the fluctuation degree of the feature in the normal state; The calculation formula of the health deviation is as follows: In the formula, represents the health deviation degree of the i-th independent abnormal region, and is used to comprehensively reflect the overall difference between the independent abnormal region and the normal region.

6. A system for assessing the health status of tomato foliage according to claim 5, characterized in that: The health deviation of all independent abnormal areas is normalized by using a preset global maximum health deviation value and a global minimum health deviation value to convert the health deviation of each independent abnormal area to the [0, 1] interval, and the mean value of the normalized health deviation of all independent abnormal areas is calculated as the mean value of the health deviation of the tomato leaf to be identified; The area ratio of the abnormal area of the tomato leaf to be identified is obtained, which is the ratio of the total pixel number of all independent abnormal areas to the total pixel number of the leaf area; According to the mean value of the health deviation and the area ratio of the abnormal area, a comprehensive health evaluation index for representing the health status of the tomato leaf is determined.

7. A system for assessing the health status of tomato foliage according to claim 6, characterized in that: The comprehensive health evaluation index is compared with a preset judgment threshold to judge the health status of the tomato leaf to be identified, and the specific logic is as follows: If then it is determined that the current tomato leaf to be identified is in a severe abnormal state; If then it is determined that the current tomato leaf to be identified is in a moderate abnormal state; If then it is determined that the current tomato leaf to be identified is in a mild abnormal state; If then the current tomato leaf to be identified is determined to be in a healthy state; In the formula, is a comprehensive health assessment index, , and is a preset determination threshold, and .

8. A method for assessing the health status of tomato foliage, characterized in that: The tomato leaf health status evaluation method adopts the tomato leaf health status evaluation system of any one of claims 1-7 to obtain, and the specific steps include: Step 1: The original image of the tomato leaf to be identified is subjected to size normalization, standardization and RGB to Lab color space conversion, a leaf area mask is generated based on a and b channel threshold, and a first image containing only the tomato leaf area is extracted; Step 2: The mean value and standard deviation of the a channel and b channel in the first image are calculated, and a sensitivity coefficient is set based on the same, pixel-level anomaly detection is performed, and abnormal area and normal area masks are generated according to the detection results; Step 3: The abnormal area mask is subjected to connected component analysis, each independent abnormal area is extracted, and its local color feature and local texture feature are extracted as identification features, and the corresponding features of the normal area are extracted as reference features; Step 4: The identification features of each independent abnormal area are compared with the reference features of the normal area item by item, the feature distance set is calculated, and the health deviation of each independent abnormal area is calculated; Step 5: The health deviation of all independent abnormal areas is normalized, and the area ratio of the abnormal area of the tomato leaf to be identified is combined to determine a comprehensive health evaluation index for representing the health status of the leaf; Step 6: According to the comprehensive health evaluation index, the health status of the tomato leaf to be identified is judged according to the preset judgment threshold rule, and the overall health evaluation result of the tomato leaf to be detected is output.

Images