Cassava leaf character and disease collaborative analysis method

Abstract

The invention provides a cassava leaf character and disease collaborative analysis method, which comprises the following steps: grouping extended form quantitative data, analyzing the density of local inter-pulse distance values in combination with leaf vein density distribution data, and dividing a high-density pulse network area and a low-density pulse network area; dividing bacterial and fungal disease potential groups based on the proportion of remarkably along-pulse extension type disease spots in each cluster of the multi-disease coexistence disease spots according to the expansion stage specificity distribution diagram, and determining the specific evolution trend of the disease species; evaluating the coincidence degree between the specific evolution trend of the disease species and the typical form of the preset bacterial disease, and when the coincidence degree is at a high level, classifying the disease species into bacterial disease groups to obtain a preliminary disease type division result; and extracting unclassified groups from the preliminary disease type division result, evaluating the similarity degree with typical morphological characteristics of fungal diseases, and outputting a bacterial and fungal disease type distinguishing result of the multi-disease coexistence scene.

Description

Technical Field

[0001] This invention relates to the field of information technology, and in particular to a method for the synergistic analysis of cassava leaf characteristics and diseases. Background Technology

[0002] Cassava, as an important food and industrial raw material crop in tropical and subtropical regions, relies heavily on accurate identification of leaf diseases, which directly impacts field management and yield assurance. In these growing areas, bacterial and fungal diseases often occur together on the same leaf, making disease type identification exceptionally difficult. Traditional identification methods primarily rely on the external morphology of lesions, particularly whether they extend clearly along the leaf veins. Typically, in bacterial vascular diseases such as bacterial black rot, the long axis of the lesions is almost parallel to the midrib and extends prominently, seemingly spreading rapidly along the leaf veins. In fungal diseases such as brown spot, lesions often spread in a circular or irregular pattern, not strictly limited by the leaf veins. However, this morphological judgment is frequently inaccurate in actual field observations, primarily due to the significant differences in vein density across different leaf regions. For example, near the midrib, where veins are thick and widely spaced, bacterial lesions can freely and rapidly expand along the veins, forming distinct, elongated stripes. In the interveinal zone of the middle or leaf margin, the veins are densely arranged with extremely small spacing, only a few millimeters wide. Even circular, spreading lesions of fungal diseases are compressed and deformed by the narrow interveinal space, presenting a false appearance of elongated extension along the veins. This illusion makes diseases belonging to different pathogens look highly similar in appearance. It is difficult to reliably distinguish between bacterial and fungal infections simply by relying on the angle between the long axis of the lesion and the main vein or the degree of extension. Among them, the method for isolating and identifying the pathogen of tobacco bacterial black spot disease published by Chinese Patent Publication No. CN113913338A on January 11, 2022, mainly relies on pathogen isolation, culture, and gene amplification methods under laboratory conditions. Although it is highly efficient in identifying pathogens of a single disease, it cannot adapt to the analysis of multiple coexisting diseases based on leaf images. It is difficult to distinguish between bacterial and fungal diseases by the interaction between lesion morphology and leaf vein structure, and it cannot predict the evolution trend of the disease. In field scenarios where multiple diseases occur simultaneously, leaves may simultaneously exhibit bacterial black rot lesions that rapidly expand along the midrib, and fungal brown spot lesions that slowly spread in the interveinal area but are compressed into elongated shapes due to the dense veins. In such cases, if only the presence of "vein-extending" lesions is considered, both lesions of different natures could be misclassified as the same type. The morphological interference caused by vein density, coupled with the shape changes of the lesions themselves at different stages of expansion, significantly reduces the reliability of the vein-extending characteristic, becoming a key obstacle to accurately distinguishing between bacterial and fungal diseases. Summary of the Invention

[0003] This invention provides a method for the synergistic analysis of cassava leaf traits and diseases, mainly including:

[0004] The original multispectral image data is retrieved from the preset cassava leaf image database, the boundary contour of the lesion area is extracted, the ratio of the major axis to the minor axis and the angle of the major axis are identified, and the initial data set of lesion morphology is obtained. Based on the initial data set of lesion morphology, the leaf vein network structure of the leaf is obtained by extracting the leaf vein skeleton, and the angle between the long axis of the lesion and the adjacent main vein and the local interveinal distance value are identified to determine the quantitative data of the expansion morphology. The quantitative data of extended morphology were grouped and combined with the leaf vein density distribution data to analyze the density of local interveinal distance values, and high-density vein network area and low-density vein network area were divided. The dispersion of the angle distribution between the long axis of the lesion sample and the direction of multiple adjacent secondary veins in the high-density vein network area is calculated. The dispersion threshold is determined based on the number and angle of the vein branches from the centroid of the lesion to the vein network node. When the dispersion is greater than the dispersion threshold, it is judged as a non-directional expansion mode. When the dispersion is less than the dispersion threshold and the angle between the long axis and the main vein is less than the preset angle threshold, the true vein-along expansion characteristics are determined based on the average vertical distance from the edge of the lesion to the nearest main vein. Based on the screening results of the actual vein extension characteristics, the proportion of lesions with significant vein extension type was analyzed, and the preliminary disease type classification results were determined.

[0005] Furthermore, raw multispectral image data is retrieved from a pre-defined cassava leaf image database, the boundary contours of lesion areas are extracted, and the ratio of the major axis to the minor axis and the angle of the major axis are identified to obtain an initial set of lesion morphology data, including: The original multispectral image was binarized to separate the lesion area from the background area of ​​healthy leaves, and a lesion candidate area mask was obtained. Based on the lesion candidate region mask map, the hole region is filled and the edge is traced to generate a closed contour curve. Adjacent contour segments are merged to obtain a set of lesion region boundary contours. For each closed contour in the set of lesion region boundary contours, fit the minimum area circumscribed ellipse, calculate the ratio of major to minor axis and the angle of major axis direction, and store the ratio of major to minor axis and the angle of major axis direction in pairs to obtain the initial data set of lesion morphology.

[0006] Furthermore, based on the initial dataset of lesion morphology, the leaf vein network structure of the leaf was obtained by extracting the leaf vein skeleton, identifying the angle between the long axis of the lesion and the adjacent midrib, as well as the local interveinal distance, to determine the quantitative data of the expansion morphology, including: Based on the lesion outline location information recorded in the initial data set of lesion morphology, the corresponding original multispectral image is obtained, and the leaf vein skeleton image is obtained by using a morphological thinning algorithm. Based on the leaf vein skeleton image, the branching nodes are identified and the main veins and secondary veins are marked to obtain a leaf vein network structure containing topological connections. For each lesion sample in the initial dataset of lesion morphology, locate its outline centroid coordinates, and measure the angle between the lesion's long axis and the adjacent main vein, as well as the local intervessel distance. The angle between the long axis of the lesion and the adjacent main vein, the local inter-vessel distance, and the ratio of the long axis to the short axis and the angle of the long axis direction of the corresponding lesion sample in the initial data set of lesion morphology are correlated to determine the quantitative data of the extended morphology.

[0007] Furthermore, the quantitative data on extended morphology were grouped, and combined with leaf vein density distribution data, the density of local interveinal distance values ​​was analyzed to divide the vein network into high-density and low-density areas, including: Based on the local interpulse distance value of each lesion sample in the extended morphological quantification data, the K-means clustering algorithm is used to group the samples into two lesion sample clusters. By comparing the average values ​​of local interpulse distances within two clusters, high-density pulse network regions and low-density pulse network regions are identified.

[0008] Furthermore, the dispersion of the angle distribution between the long axis of the lesion sample and the direction of multiple adjacent secondary veins within the high-density vein network area is calculated. A dispersion threshold is determined based on the number and angle of vein branches from the centroid of the lesion to the vein network node. When the dispersion is greater than the dispersion threshold, it is determined to be a non-directional expansion pattern. When the dispersion is less than the dispersion threshold and the angle between the long axis and the main vein is less than a preset angle threshold, the true vein-along expansion characteristics are determined based on the average vertical distance from the lesion edge to the nearest main vein, including: For lesion samples carrying high-density vein network area attribution markers, secondary leaf vein branches within a preset radius around the centroid of the lesion outline are extracted, the angle between the long axis of the lesion and the direction of each secondary leaf vein is measured, and the standard deviation of the angle numerical sequence is calculated as the angle distribution dispersion. Based on the centroid coordinates of the lesion outline, locate the nearest leaf vein bifurcation node, count the number of leaf vein branches extending from the bifurcation node, and determine the corresponding dispersion threshold based on the number of leaf vein branches. When the dispersion of the included angle distribution is greater than the dispersion threshold, the lesion sample is marked as a non-directional expansion mode; when the dispersion of the included angle distribution is less than or equal to the dispersion threshold and the angle between the long axis of the lesion and the adjacent main vein is less than a preset angle threshold, multiple points are sampled along the edge of the lesion contour, the average vertical distance from each point to the nearest main vein is calculated, and the true vein-along-line expansion feature is determined and added based on the average vertical distance.

[0009] Furthermore, based on the screening results of the actual vein-extending characteristics, the proportion of lesions with significant vein-extending characteristics was analyzed, and the preliminary disease type classification results were determined, including: Based on the screening results of the actual vein extension characteristics, when the ratio of the major axis to the minor axis of the lesion sample in the low-density vein network area is greater than the threshold and the angle between the major axis and the main vein is less than the preset angle, it is marked as a significant vein extension type. The judgment results of the high-density vein network area and the low-density vein network area are merged to generate an anisotropic distribution map of the extension stage. Based on the heterogeneous distribution map of the expansion stage, the potential groups of bacterial and fungal diseases are divided according to the proportion of lesions with significant vein-extending types in each cluster of lesions with multiple coexisting diseases, and the disease-specific evolution trend is determined. Assess the degree of agreement between the disease-specific evolution trend and the pre-defined typical morphology of bacterial diseases. When the degree of agreement is high, classify the disease into the bacterial disease group to obtain preliminary disease type classification results.

[0010] Furthermore, based on the screening results of the actual vein-wide extension characteristics, when the aspect ratio of the lesion sample in the low-density vein network area is greater than the threshold and the angle between the long axis and the main vein is less than the preset angle, it is marked as a significant vein-wide extension type. The determination results of the high-density vein network area and the low-density vein network area are merged to generate an anisotropic distribution map of the extension stage, including: Based on the true vein-along extension feature identifiers, lesion samples carrying true vein-along extension feature identifiers are extracted from high-density vein network areas and labeled as vein-along extension types, while non-directional extension pattern identifiers are retained; For lesion samples in low-density venous network areas, they are marked as either significantly extended along the vein or ordinary extended based on the ratio of the major axis to the minor axis and the angle between the major axis and the main vein. The lesion expansion type records of high-density venous network area and low-density venous network area are merged, and the type labels are mapped based on the outline centroid coordinates to generate an anisotropic distribution map of the expansion stage.

[0011] Furthermore, based on the heterogeneous distribution map of the expansion stage, and according to the proportion of lesions with significant vein-extending characteristics within each cluster of lesions coexisting with multiple diseases, potential groups of bacterial and fungal diseases were classified, and the disease-specific evolutionary trends were determined, including: Based on the centroid coordinates of the lesion outlines in the heterogeneous distribution map of the extended stage, a hierarchical clustering algorithm is used to group the lesion samples to obtain multiple lesion clusters; For each lesion cluster, the proportion of lesions with significant vein extension is counted. When the proportion is greater than a preset threshold, the cluster is classified into the potential bacterial disease group; otherwise, it is classified into the potential fungal disease group. Based on the classification results, the disease-specific evolution trend extending along the main vein is marked for potential bacterial disease groups, and the disease-specific evolution trend scattered in the intervessel region is marked for potential fungal disease groups, thus determining the disease-specific evolution trend.

[0012] Furthermore, the degree of agreement between the disease-specific evolution trend and the pre-defined typical morphology of bacterial diseases is assessed. When the degree of agreement is high, the disease is classified into the bacterial disease group, obtaining preliminary disease type classification results, including: Extract the standard ranges for the ratio of major axis to minor axis, the standard range for the angle of the main vein, and the standard range for the proportion of the types of vein extension from the pre-defined typical morphological descriptions of bacterial diseases. For each potential bacterial disease group, the average ratio of major axis to minor axis, the average angle of the main vein, and the percentage of vein extension types were statistically analyzed for lesion clusters. The statistical results are compared with the standard range item by item, and the proportion of feature items falling within the standard range is calculated as the degree of consistency value. When the degree of consistency value is greater than the preset degree of consistency threshold, the cluster is marked as a bacterial disease group, and the preliminary disease type classification results are obtained by summarizing.

[0013] Furthermore, after determining the preliminary disease type classification results, this includes: Clusters of lesions that retain potential group markers are extracted from the preliminary disease type classification results and classified as unclassified groups. For lesion clusters within potential fungal disease groups, the proportion of non-vein-spreading lesions is calculated. When the proportion is greater than a preset fungal proportion threshold, it is marked as a fungal disease group. It summarizes bacterial disease groups, fungal disease groups, and clusters with retained potential group markers, records their spatial distribution and final classification status, and outputs the results of distinguishing between bacterial and fungal disease types.

[0014] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects: This invention discloses a method for the collaborative analysis of cassava leaf traits and diseases. It extracts the lesion boundary contours, their aspect ratios, and the direction of the long axis from multispectral images. Combined with the network structure obtained from the leaf vein skeleton, it quantifies the angle between the long axis of the lesion and the adjacent midrib, as well as the local interveinal distance. Then, a clustering algorithm is used to divide the vein network into high- and low-density areas. Based on this, the dispersion of the angle between the long axis of the lesion and the secondary vein in the high-density area is calculated. Non-directional expansion or true vein-along extension characteristics are determined using dispersion thresholds and midrib angle thresholds. Finally, lesions with significant vein-along extension are screened, generating an anisotropic distribution map of the expansion stage. Based on the differences in the proportion of significant vein-along extension among lesion clusters in this map, and by integrating the morphological evolution patterns under multiple coexisting diseases, it achieves accurate differentiation and classification of bacterial and fungal diseases. This effectively solves the problem that traditional single morphological features are insufficient for accurately identifying pathogen types in mixed infections of multiple diseases, improving the reliability of early accurate diagnosis and control of cassava diseases. Attached Figure Description

[0015] Figure 1This is a flowchart of a method for the synergistic analysis of cassava leaf traits and diseases according to the present invention.

[0016] Figure 2 This is a schematic diagram of a method for the synergistic analysis of cassava leaf traits and diseases according to the present invention.

[0017] Figure 3 This is another schematic diagram of a method for the synergistic analysis of cassava leaf characteristics and diseases according to the present invention. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0019] like Figures 1-3 This embodiment of a method for the synergistic analysis of cassava leaf traits and diseases may specifically include: S101. Retrieve original multispectral image data from the preset cassava leaf image database, extract the boundary contour of the lesion area, identify the ratio of the major and minor axes of the lesion contour and the angle of the major axis direction, and obtain the initial data set of lesion morphology.

[0020] Raw multispectral image data is retrieved from a pre-defined cassava leaf image database. For each frame, a grayscale segmentation threshold is calculated using the maximum inter-class variance method. The image is then binarized based on this threshold to separate the lesion area from the healthy leaf background area, resulting in a lesion candidate region mask. Based on this lesion candidate region mask, morphological closing operations are used to fill the pores within the lesions. A closed contour curve is generated pixel-by-pixel along the edge of the mask after the closing operation. If multiple discrete contour segments exist within a single lesion area, the geometric centroid coordinates of each segment are calculated. When the distance between the centroids of adjacent segments is less than a pre-defined pixel distance threshold, the contours are merged to obtain a set of lesion area boundary contours. For each closed contour in the set of boundary contours of the lesion area, a minimum area circumscribed ellipse is fitted based on the coordinates of the contour points. The lengths of the major and minor axes of the circumscribed ellipse are measured. The ratio of the major axis to the minor axis is obtained by dividing the length of the major axis by the length of the minor axis. The angle of the major axis direction relative to the horizontal baseline of the image is measured to obtain the angle value of the major axis direction. The ratio of the major axis to the minor axis and the angle value of the major axis direction are paired and stored to obtain the initial data set of lesion morphology.

[0021] In the process of identifying diseases in cassava leaves, the acquisition of multispectral image data is fundamental to the analysis of lesion morphology. A pre-set cassava leaf image database stores leaf images of different growth stages and disease types, with each image frame containing reflectance information from multiple spectral bands. When retrieving raw multispectral image data from this database, the target image sequence to be analyzed is obtained by indexing and searching according to acquisition time, leaf number, or disease label category.

[0022] In one embodiment, when performing grayscale segmentation on the retrieved multispectral image data, the maximum inter-class variance (MOL) method is used to calculate the grayscale segmentation threshold. The basic principle of this method is to divide image pixels into foreground and background classes based on their grayscale values. By iterating through all possible grayscale thresholds, the MOL between foreground and background pixels is calculated at each threshold, and the grayscale value corresponding to the maximum MOL is selected as the segmentation threshold.

[0023] Specifically, assuming the image grayscale range is 0 to 255, for each candidate threshold, the percentage of pixels below the threshold, the percentage of pixels above the threshold, and the average grayscale value of the two classes are statistically analyzed, and then the variance between the two classes is calculated. When the inter-class variance is the largest, it indicates that the difference between the foreground and background is most significant, and the threshold at this point can effectively separate the lesion area from the background area of ​​healthy leaves. After binarizing the image based on the grayscale segmentation threshold, the lesion area appears as a white foreground, and the healthy area appears as a black background, thus obtaining a lesion candidate area mask map.

[0024] It should be noted that the lesion candidate region mask may contain small holes inside the lesions or burr noise at the edges. Morphological closing operations, through a sequence of dilation followed by erosion, can fill the holes inside the lesion region while maintaining the overall outline shape of the lesion. The dilation operation uses a structuring element of a preset size to slide on the mask, setting all positions with foreground pixels within the structuring element's coverage area as foreground, thereby expanding the lesion boundary and filling small holes. The subsequent erosion operation uses a structuring element of the same size, retaining the position as foreground only when the structuring element completely covers the foreground area, thus causing the dilated boundary to shrink back to its original position.

[0025] For example, when performing contour tracing along the edge of the mask after the closing operation, the scan starts from the top left corner of the mask image and proceeds line by line. When the first foreground pixel is detected, the coordinates of adjacent boundary pixels are recorded sequentially in a clockwise direction, starting from that pixel, until the starting point is returned to form a closed contour curve. If a single lesion region produces multiple discrete contour segments due to uneven lesion development or image noise interference, the geometric centroid coordinates of each segment are calculated. The x-coordinate of the geometric centroid is the arithmetic mean of the x-coordinates of all boundary pixels of that segment, and the y-coordinate is the arithmetic mean of the y-coordinates of all boundary pixels. When the Euclidean distance between the centroids of two adjacent segments is less than a preset pixel distance threshold, the two segments are determined to belong to the same lesion, and their contour points are merged into a complete closed contour, thereby obtaining the set of boundary contours of the lesion region.

[0026] In one possible implementation, when fitting an ellipse to each closed contour in the set of contours representing the lesion region boundary, a minimum area circumscribed ellipse method is used. This method finds an ellipse that completely encloses all contour points and has the smallest area, based on the set of contour point coordinates. During the fitting process, the geometric centroid of the contour point set is used as the initial estimated position of the ellipse center. The projection width of the contour point set in different directions is determined using a rotating caliper algorithm. The two orthogonal directions with the largest and smallest projection widths are selected as the directions of the major and minor axes of the ellipse, respectively. Then, the lengths of the major and minor axes are determined based on the extreme coordinates of the contour points in these two directions.

[0027] Specifically, after measuring the lengths of the major and minor axes of the circumscribed ellipse, the ratio of the major axis to the minor axis is obtained by dividing the major axis length by the minor axis length. This value reflects the elongation or flatness of the lesion outline. Simultaneously, the angle between the major axis direction and the horizontal baseline of the image is measured, with the horizontal direction to the right of the image as the zero-degree reference and the counter-clockwise direction as the positive angle, to obtain the angle value of the major axis direction. The major-minor axis ratio value and the major axis direction angle value corresponding to each closed contour are paired and stored to form an initial data set of lesion morphology.

[0028] S102. Based on the initial data set of lesion morphology, the leaf vein network structure of the leaf is obtained by extracting the leaf vein skeleton, and the angle between the long axis of the lesion and the adjacent main vein and the local intervessel distance value are identified to determine the quantitative data of the expansion morphology.

[0029] Based on the lesion outline location information, major-minor axis ratio, and major axis angle recorded in the initial lesion morphology dataset, the corresponding original multispectral image is obtained. A leaf grayscale image is then converted from the near-infrared channel of the multispectral image. A morphological thinning algorithm is used to iteratively erode the leaf grayscale image, peeling away the edge pixels of the veins layer by layer until the vein width shrinks to a single pixel width, resulting in a vein skeleton image. Based on the vein skeleton image, positions with more than two pixel connections in the skeleton segments are identified as bifurcation nodes. Continuous skeleton segments between adjacent bifurcation nodes are marked as independent vein branches. The length of the line connecting the start and end points of each vein branch is measured. If the line length exceeds a preset length threshold of 20 pixels and is located in the longitudinal region of the central third of the leaf width, it is marked as a main vein. The remaining vein branches are marked as secondary veins, thus obtaining a vein network structure containing the topological connections between main and secondary veins. For each lesion sample in the initial dataset of lesion morphology, its centroid coordinates are located. The nearest main vein branch to this centroid is searched within the leaf vein network structure. The angle between the lesion's long axis direction vector and the direction vector of the adjacent main vein is measured to obtain the angle value between the lesion's long axis and the adjacent main vein. The vertical distance from the edge of the lesion's outline to the nearest secondary vein skeleton lines on both sides is measured, with the vertical direction reference being perpendicular to the lesion's long axis. A point is sampled every 5 pixels and averaged to obtain the local intervein distance value. The angle value between the lesion's long axis and the adjacent main vein, the local intervein distance value, are correlated and paired with the ratio of the major axis to the minor axis and the long axis direction angle value of the corresponding lesion sample in the initial dataset of lesion morphology. All paired records of lesion samples are summarized to determine the extended morphological quantification data.

[0030] In the process of identifying diseases on cassava leaves, the extraction of the leaf vein network structure is the foundation for establishing the spatial relationship between lesions and leaf veins. The initial dataset of lesion morphology records the outline centroid coordinates of each lesion. This coordinate information can be used to locate the original multispectral image region where the lesion is located, thereby obtaining a grayscale image of the leaf containing leaf vein texture information.

[0031] In one embodiment, the morphological thinning algorithm is implemented by iteratively eroding the edge pixels of the leaf veins.

[0032] Specifically, the algorithm scans the grayscale image of the leaf using structuring elements. In each iteration, it determines whether the current foreground pixel is an edge pixel, based on whether both foreground and background pixels exist within its eight neighbors. If the current pixel is an edge pixel and its removal would not disrupt the leaf vein connectivity, it is marked as a pixel to be deleted. After one round of scanning, all pixels to be deleted are deleted, and the next iteration begins. This iteration process continues until no edge pixels satisfying the deletion criteria remain in the image. At this point, the leaf vein width has shrunk to a single pixel width, forming a leaf vein skeleton image.

[0033] It should be noted that identifying bifurcation nodes in the leaf vein skeleton image is a crucial step in constructing the leaf vein network structure. A bifurcation node is defined as a position in the skeleton line segment where the number of pixel connections is greater than two, meaning that the pixel is directly connected to three or more adjacent skeleton pixels. In the eight-neighbor connectivity determination, each foreground pixel in the skeleton image is examined in its eight surrounding directions, and the number of adjacent foreground pixels is counted. When the number exceeds two, the position is marked as a bifurcation node.

[0034] For example, after identifying all bifurcation nodes, the continuous skeleton pixel sequence between two adjacent bifurcation nodes is extracted as an independent leaf vein branch. The extraction process starts from a certain bifurcation node and traces adjacent pixels along the skeleton line segment until the next bifurcation node or the end of the skeleton is encountered. All pixels on the tracing path constitute a leaf vein branch. For cassava leaves, the main vein is usually located in the central longitudinal region of the leaf and is relatively long, while secondary veins branch off from both sides of the main vein and are relatively short. Therefore, the length of the line connecting the start and end points of each leaf vein branch is measured. If the length exceeds a preset length threshold and the branch is located in the central longitudinal region of the leaf, the branch is marked as a main vein; branches that do not meet the above conditions are marked as secondary veins. The connections between all leaf vein branches and their bifurcation nodes together constitute the leaf vein network structure.

[0035] In one possible implementation, the measurement of the angle between the long axis of the lesion and the adjacent midrib depends on the spatial correspondence between the centroid of the lesion outline and the leaf vein network structure. For each lesion sample in the initial lesion morphology dataset, the horizontal and vertical coordinates of its centroid are read. Then, all midrib branches are traversed in the leaf vein network structure, and the shortest distance from the centroid coordinates to the skeleton line of each midrib branch is calculated. The midrib branch corresponding to the shortest distance is selected as the adjacent midrib of the lesion. The direction vector of the lesion's long axis is determined by the angle value of the long axis recorded in the initial lesion morphology dataset, and the direction vector of the adjacent midrib is determined by the direction from the starting point to the ending point of the midrib branch. The angle between the two vectors is calculated using the vector dot product formula. This angle value is the angle between the long axis of the lesion and the adjacent midrib, and its value ranges from 0 degrees to 90 degrees.

[0036] Specifically, the measurement of local interveinal distance reflects the density of leaf vein distribution in the lesion area. Using the lesion outline edge pixels as a reference, the nearest secondary vein skeleton line is searched to the left and right, respectively, with the search direction perpendicular to the long axis of the lesion. When the search ray intersects a secondary vein skeleton line, the Euclidean distance between the intersection point and the lesion outline edge pixels is recorded as the one-sided interveinal distance. This measurement process is repeated for multiple sampling points on the lesion outline. The arithmetic mean of the left and right interveinal distances is calculated, and the two averages are added together and divided by two to obtain the local interveinal distance value for that lesion sample.

[0037] Preferably, the process of forming extended morphological quantification data involves integrating the quantification results of the spatial relationship between lesions and leaf veins with the morphological parameters of the lesions themselves. For each lesion sample, the ratio of its major axis to minor axis and the angle of its major axis direction are extracted from the initial lesion morphology data set, and then correlated and paired with the previously measured angle between the lesion's major axis and the adjacent midrib, and the local interveinal distance, to form a lesion morphology description record containing four dimensions.

[0038] Understandably, compiling the four-dimensional morphological descriptions of all lesion samples yields complete extended morphological quantitative data. Each record in this data corresponds to one lesion sample, and the record content covers the geometric morphological features of the lesion outline and the spatial relationship between the lesion and the surrounding leaf vein network. This provides multi-dimensional quantitative evidence for distinguishing between morphological artifacts caused by differences in leaf vein density and true vein-along extension characteristics.

[0039] S103. Group the quantitative data of extended morphology, combine it with the leaf vein density distribution data, analyze the density of local interveinal distance values, and divide the high-density vein network area and the low-density vein network area.

[0040] Based on the local interveinal distance values ​​of each lesion sample in the extended morphological quantification data, the K-means clustering algorithm was used to group all lesion samples. The number of clusters was set to two, and the local interveinal distance value was used as the clustering feature dimension. The distance from each sample to the cluster center was iteratively calculated and the cluster center position was updated until the cluster center position no longer changed, resulting in two lesion sample clusters. For each of the two lesion sample clusters, the arithmetic mean of the local interveinal distance values ​​of all lesion samples within each cluster was calculated. The average values ​​of the two clusters were compared. Since a smaller local interveinal distance value indicates a denser vein distribution in that area, the cluster with the smaller average value was labeled as a high-density vein network area, and the cluster with the larger average value was labeled as a low-density vein network area, thus obtaining the vein network area type label for each cluster. Based on the pulse network region type label, a high-density pulse network region attribution identifier is added to each lesion sample in the cluster labeled as a high-density pulse network region, and a low-density pulse network region attribution identifier is added to each lesion sample in the cluster labeled as a low-density pulse network region, thus completing the division of high-density pulse network regions and low-density pulse network regions.

[0041] In the process of identifying diseases on cassava leaves, the distribution density of leaf veins varies in different leaf regions, and this difference directly affects the presentation of lesion expansion morphology. The local interveinal distance value recorded in the quantitative data of expansion morphology reflects the density of leaf veins in each lesion area. By clustering this value, lesion samples can be classified according to the density characteristics of their surrounding leaf vein environment.

[0042] In one embodiment, the K-means clustering algorithm comprises two phases: initialization and iterative update. In the initialization phase, two values ​​are randomly selected from the local interpulse distances of all lesion samples as initial cluster centers. In the iterative update phase, two operations are executed sequentially: sample allocation and cluster center update. During sample allocation, the absolute value of the difference between the local interpulse distance of each lesion sample and the two cluster centers is calculated, and the sample is assigned to the cluster represented by the cluster center with the smaller absolute difference. During cluster center update, the arithmetic mean of the local interpulse distances of all lesion samples within each cluster is calculated, and this mean is used as the new cluster center. The sample allocation and cluster center update operations are repeated until the difference between the cluster center positions of two consecutive iterations is less than a preset convergence threshold, at which point the clustering process ends, resulting in two lesion sample clusters.

[0043] It should be noted that there is an inverse relationship between the local interveinal distance value and the leaf vein density. The local interveinal distance value represents the average vertical distance from the edge of the lesion outline to the skeleton line of the nearest secondary leaf vein on both sides. The smaller the value, the closer the secondary leaf veins on both sides of the lesion are to the lesion, that is, the denser the leaf vein distribution in this area; conversely, the larger the value, the farther the secondary leaf veins on both sides of the lesion are from the lesion, that is, the sparser the leaf vein distribution in this area.

[0044] Specifically, when comparing the average values ​​of two clusters, the average interveinal distance of all lesion samples within each cluster is summed and divided by the number of samples to obtain the average interveinal distance of that cluster. Clusters with smaller average values ​​correspond to leaf areas with dense vein distribution and are therefore labeled as high-density vein network areas; clusters with larger average values ​​correspond to leaf areas with sparse vein distribution and are therefore labeled as low-density vein network areas.

[0045] In one possible implementation, the addition of vein network region attribution identifiers uses a binary label format. For each lesion sample within a cluster labeled as a high-density vein network region, a high-density vein network region attribution identifier field is appended to its extended morphological quantification data record; similarly, for each lesion sample within a cluster labeled as a low-density vein network region, a low-density vein network region attribution identifier field is appended to its extended morphological quantification data record. After these operations, each lesion sample carries clear vein network region attribution information, thus achieving the distinction between high-density and low-density vein network regions.

[0046] S104. Calculate the angular distribution dispersion of the long axis of the lesion sample and the direction of multiple adjacent secondary leaf veins in the high-density vein network area. Determine the dispersion threshold based on the number and angle of the leaf vein branches from the centroid of the lesion to the leaf vein network node. When the dispersion is greater than the dispersion threshold, it is judged as a non-directional expansion mode. When the dispersion is less than the dispersion threshold and the angle between the long axis and the main vein is less than the preset angle threshold, determine the true vein-along expansion characteristics based on the average vertical distance from the edge of the lesion to the nearest main vein.

[0047] For lesion samples carrying high-density vein network area attribution identifiers, all secondary vein branches within a preset radius around the centroid of the lesion outline are extracted from the leaf vein network structure. The angle between the direction vector of the lesion's long axis and the direction vector of each secondary vein is measured, and all angle values ​​are compiled into an angle value sequence. The standard deviation of the angle value sequence is calculated as the angle distribution dispersion. Based on the coordinates of the lesion outline centroid, the leaf vein bifurcation node closest to the centroid is located in the leaf vein network structure. The number of leaf vein branches extending from this bifurcation node is counted. When the number of leaf vein branches exceeds a preset branch number threshold, a higher dispersion threshold is used; when the number of leaf vein branches is less than or equal to the preset branch number threshold, a lower dispersion threshold is used. This determines the dispersion threshold corresponding to the lesion sample. If the angle distribution dispersion is greater than the dispersion threshold, the lesion sample is marked as a non-directional expansion pattern. If the angle distribution dispersion is less than or equal to the dispersion threshold, and the angle between the lesion's long axis and the adjacent main vein recorded in the expansion morphology quantification data is less than a preset angle threshold, the process proceeds to the vein expansion feature determination process. For lesion samples that enter the vein extension feature determination process, multiple edge points are uniformly sampled along the edge of the lesion contour. The vertical distance from each edge point to the nearest main vein skeleton line is measured. The arithmetic mean of the vertical distances of all edge points is calculated as the mean vertical distance from the edge to the main vein. If the mean vertical distance from the edge to the main vein is less than a preset distance threshold, a true vein extension feature identifier is added to the lesion sample.

[0048] Within high-density vein network areas, the morphology of lesion expansion is easily deformed due to the spatial constraints of dense leaf veins. Therefore, multi-angle quantitative analysis is needed to determine whether the actual expansion direction of the lesion truly follows the leaf vein direction. Lesion samples carrying high-density vein network area identification marks are the subjects of this step, and the spatial relationship between the long axis direction of these samples and the direction of surrounding secondary leaf veins is quantitatively described.

[0049] In one embodiment, the extraction range of secondary vein branches is centered on the centroid of the lesion outline, with a preset radius covering the local vein network area surrounding the lesion. All secondary vein branches falling within this circular range are selected from the vein network structure. The direction vector of each secondary vein branch is determined by the direction from its starting point to its ending point. The long axis direction vector of the lesion is calculated from the long axis angle value recorded in the extended morphology quantification data. When measuring the angle between the long axis direction vector of the lesion and the direction vector of each secondary vein, the cosine value of the angle between the two vectors is calculated using the vector dot product formula, and then the angle value is obtained using the inverse cosine function. The angle values ​​between the lesion and all secondary vein branches are arranged sequentially to form an angle value sequence.

[0050] It should be noted that the dispersion of the angle distribution is characterized by the standard deviation of the angle value sequence. The standard deviation is calculated by first obtaining the arithmetic mean of the angle value sequence, then calculating the difference between each angle value and the mean, summing the squares of all differences, dividing by the number of angles, and finally taking the square root of the quotient. The larger the dispersion value of the angle distribution, the greater the difference in the angle between the long axis of the lesion and the direction of the surrounding secondary veins, meaning that the long axis of the lesion does not significantly deviate from any particular vein direction; the smaller the dispersion value of the angle distribution, the more consistent the angle between the long axis of the lesion and the direction of the surrounding secondary veins, meaning that the long axis of the lesion may extend along a certain vein direction.

[0051] Specifically, a leaf vein bifurcation node is the location where leaf vein branches intersect in the leaf vein network structure. When locating the leaf vein bifurcation node closest to the centroid of the lesion outline, all bifurcation nodes in the leaf vein network structure are traversed, the Euclidean distance between the coordinates of each bifurcation node and the coordinates of the centroid of the lesion outline is calculated, and the bifurcation node with the smallest Euclidean distance is selected as the target bifurcation node.

[0052] In one possible implementation, the dispersion threshold is determined in relation to the number of vein branches extending from the target bifurcation node. The number of vein branches reflects the complexity of the vein network in that region; more branches mean more dispersed vein orientation, and a greater likelihood that lesions are constrained by veins from different directions. When the number of vein branches exceeds a preset branch number threshold, it indicates a high complexity of the vein network in that region. Even if the lesion does not extend along the veins, it may exhibit a large angular dispersion. In this case, a higher dispersion threshold is used to improve the judgment criteria. When the number of vein branches is less than or equal to the preset branch number threshold, it indicates a low complexity of the vein network in that region. The lesion's expansion direction is less affected by interference. In this case, a lower dispersion threshold is used to maintain judgment sensitivity.

[0053] For example, the determination of non-directional expansion patterns is based on the comparison between the dispersion of the angle distribution and the dispersion threshold. When the dispersion of the angle distribution is greater than the dispersion threshold, it indicates that there is no obvious directional consistency between the long axis of the lesion and the direction of the surrounding secondary leaf veins. The morphology of the lesion is more likely to be a random deformation formed by compression from leaf veins in multiple directions, rather than a directional expansion along a specific leaf vein direction. Therefore, the lesion sample is marked as a non-directional expansion pattern. Furthermore, when the dispersion of the angle distribution is less than or equal to the dispersion threshold, it is also necessary to combine the angle between the long axis of the lesion and the adjacent main vein for a comprehensive judgment. If the angle between the long axis of the lesion and the adjacent main vein recorded in the expansion morphology quantification data is less than the preset angle threshold, it indicates that the direction of the long axis of the lesion is relatively close to the direction of the main vein. The lesion may have the characteristic of expanding along the main vein. At this time, the vein-expansion characteristic determination process is entered for more refined verification.

[0054] Preferably, the vein-wide extension feature determination process verifies whether a lesion is closely attached to the main vein by measuring the vertical distance from the lesion edge to the main vein. Multiple edge points are uniformly sampled along the lesion contour edge at fixed angular intervals. For each edge point, the nearest skeleton pixel on the main vein skeleton line is searched, and the Euclidean distance between the edge point coordinates and the skeleton pixel coordinates is calculated as the vertical distance from the edge point to the main vein. The arithmetic mean of the vertical distances of all edge points is then calculated to obtain the average vertical distance from the edge to the main vein.

[0055] Understandably, the average vertical distance from the edge to the main vein reflects the proximity of the overall outline of the lesion to the main vein. If this average is less than a preset distance threshold, it indicates that the edge of the lesion outline is generally close to the main vein skeleton line, and the spatial distribution of the lesion shows a morphological characteristic of extending along the main vein direction. In this case, adding a true vein-along extension feature marker to the lesion sample indicates that the slender shape exhibited by the lesion in the high-density vein network area is a true vein-along extension behavior rather than an illusion caused by leaf vein compression.

[0056] S105. Based on the screening results of the actual vein extension characteristics, when the ratio of the major axis to the minor axis of the lesion sample in the low-density vein network area is greater than the threshold and the angle between the major axis and the main vein is less than the preset angle, it is marked as a significant vein extension type. The judgment results of the high-density vein network area and the low-density vein network area are merged to generate an anisotropic distribution map of the extension stage.

[0057] Based on the screening results of the true vein-along extension feature identifiers, lesion samples carrying the true vein-along extension feature identifiers are extracted from the high-density vein network area and labeled as vein-along extension type. Lesion samples labeled as non-directional extension mode are extracted and their non-directional extension type labels are retained. The contour centroid coordinates and corresponding type labels of the two types of lesion samples are recorded to obtain the lesion extension type record set of the high-density vein network area. For lesion samples carrying low-density vein network area attribution identifiers, the major-minor axis ratio and the angle between the lesion's major axis and the adjacent main vein are read from the extension morphology quantification data for each lesion sample. If the major-minor axis ratio is greater than a preset major-minor axis ratio threshold and the angle between the lesion's major axis and the adjacent main vein is less than a preset angle threshold, a significant vein-along extension type label is added to the lesion sample; otherwise, a normal extension type label is added to the lesion sample. The contour centroid coordinates and corresponding type labels of all lesion samples are recorded to obtain the lesion extension type record set of the low-density vein network area. The high-density vein network area lesion expansion type record set and the low-density vein network area lesion expansion type record set are merged. Using the leaf image coordinates as a reference, the outline centroid coordinates of each lesion sample and the corresponding type label are mapped to the leaf spatial location to generate an anisotropic distribution map of the expansion stage.

[0058] After determining the expansion characteristics of lesion samples in the high-density venous network region, it is necessary to classify the expansion types of lesion samples in the low-density venous network region accordingly, and integrate the determination results of the two venous network regions to form a complete spatial distribution map. The establishment of the lesion expansion type record set in the high-density venous network region is based on the screening results of the previous steps. Lesion samples carrying true venous expansion characteristic identifiers are classified as venous expansion types, while lesion samples marked as non-directional expansion patterns retain their original type labels.

[0059] In one embodiment, the expansion type determination of lesion samples within the low-density venous network region employs a dual-condition joint judgment method. The major-minor axis ratio of each lesion sample is read from the expansion morphology quantification data. This value reflects the elongation of the lesion outline; a larger value indicates a narrower and more elongated lesion shape. Simultaneously, the angle between the lesion's major axis and the adjacent main vein is read. This value reflects the degree of deviation between the lesion's major axis direction and the main vein's direction; a smaller value indicates that the lesion's major axis is closer to being parallel to the main vein. When the major-minor axis ratio is greater than a preset major-minor axis ratio threshold and the angle between the lesion's major axis and the adjacent main vein is less than a preset angle threshold, it indicates that the lesion exhibits both a distinctly elongated shape and a high degree of consistency with the main vein's direction, conforming to the typical characteristics of vein-associated expansion. A significant vein-associated extension type marker is then added to this lesion sample.

[0060] It should be noted that for low-density vascular network lesion samples that do not meet the above two conditions, a common extension type marker is added. These lesions may have a relatively round shape, or although they are elongated, their long axis deviates significantly from the direction of the main vein, thus lacking the characteristic of significant extension along the vein.

[0061] Specifically, the merging process of the two lesion expansion type record sets in the vein network area used the pixel coordinates of the original leaf image as a unified reference. The high-density vein network area lesion expansion type record set contained two types of lesion samples: those extending along the veins and those extending in a non-directional manner. The low-density vein network area lesion expansion type record set contained two types of lesion samples: those extending significantly along the veins and those extending in a normal manner. All lesion samples in the two record sets were located according to the centroid coordinates of their outlines, and the location point of each lesion sample and its corresponding type label were marked within the leaf image space.

[0062] For example, the heterogeneous distribution map of the expansion stage uses the leaf outline as the boundary, visually presenting the spatial distribution of various types of lesion samples in different areas of the leaf. Lesion samples of the vein-spreading type and the significantly vein-extending type represent lesions with expansion characteristics along the leaf vein direction, while lesion samples of the non-directional expansion type and the ordinary expansion type represent lesions without obvious directional expansion characteristics. This distribution map uniformly presents the determination results of high-density and low-density vein network areas, reflecting the heterogeneous distribution pattern of lesion expansion characteristics in leaf space.

[0063] S106. Based on the heterogeneous distribution map of the expansion stage, and based on the proportion of lesions with significant vein-extending types in each cluster of lesions with multiple coexisting diseases, the potential groups of bacterial and fungal diseases are divided, and the disease-specific evolution trend is determined.

[0064] Based on the centroid coordinates of the outlines of each lesion sample in the heterogeneous distribution map of the expansion stage, a hierarchical clustering algorithm is used to group all lesion samples according to spatial Euclidean distance. Lesion samples with a spatial distance less than a preset clustering distance threshold are grouped into the same lesion cluster, resulting in multiple lesion clusters and a list of lesion samples they contain. For each lesion cluster, the number of lesion samples carrying the vein extension type marker and the number of lesion samples carrying the significant vein extension type marker are counted. The ratio of the sum of these two to the total number of lesion samples in the cluster is calculated to obtain the vein extension type proportion value. If the vein extension type proportion value is greater than a preset proportion threshold, the lesion cluster is classified into the potential bacterial disease group; otherwise, the lesion cluster is classified into the potential fungal disease group. Based on the classification results of the potential groups of bacterial diseases and fungal diseases, the disease-specific evolution trend of lesion clusters extending along the main vein direction is marked for the bacterial disease potential group, and the disease-specific evolution trend of lesion clusters scattered in the intervessel region is marked for the fungal disease potential group, thus determining the disease-specific evolution trend.

[0065] In scenarios where multiple diseases coexist, lesions from both bacterial and fungal diseases often appear on the same leaf, exhibiting a clustered spatial distribution. The heterogeneous distribution map of the expansion stage records the outline centroid coordinates of each lesion sample and its corresponding type label. Based on this information, spatial clustering of lesion samples can identify lesion clusters with similar spatial distribution characteristics.

[0066] In one embodiment, the hierarchical clustering algorithm is implemented using a bottom-up agglomerative approach. Initially, the centroid coordinates of each lesion sample's outline are considered as an independent clustering unit. During clustering, the spatial Euclidean distance between any two clustering units is calculated, and the two clustering units with the smallest distance are merged to form a new clustering unit. This distance calculation and unit merging operation is repeated until the minimum distance between all clustering units is greater than a preset clustering distance threshold, at which point the clustering process terminates. At this point, the lesion samples contained within each clustering unit constitute a lesion cluster, and these lesions exhibit a close proximity distribution in the leaf space.

[0067] It should be noted that the calculation of the proportion of lesion extension types comprehensively considers the lesion extension characteristics from both high-density and low-density lesion network areas. Lesion samples carrying lesion extension type markers within high-density lesion network areas underwent rigorous dispersion determination and edge distance verification, while lesion samples carrying significant lesion extension type markers within low-density lesion network areas underwent dual screening based on the major-minor axis ratio and the main vein angle. The proportion of lesion extension types obtained by adding the number of these two types of lesion samples and dividing by the total number of lesion samples within the cluster reflects the overall lesion extension tendency of the cluster.

[0068] Specifically, the classification of potential bacterial and fungal diseases into groups is based on a comparison of the proportion of lesion patterns extending along the veins with a preset threshold. When the proportion of lesion patterns extending along the veins in a cluster is greater than the preset threshold, it indicates that most lesions in the cluster exhibit the characteristic of extending along the leaf veins, consistent with the biological characteristics of rapid spread of bacterial vascular diseases along the xylem tissue. Therefore, this cluster of lesions is classified into the potential bacterial disease group. When the proportion of lesions extending along the veins is less than or equal to the preset threshold, it indicates that the lesions in the cluster mainly exhibit non-directional or general expansion characteristics, consistent with the biological characteristics of fungal diseases spreading radially or irregularly in leaf tissue. Therefore, this cluster of lesions is classified into the potential fungal disease group.

[0069] For example, the determination of disease-specific evolutionary trends is based on the spatial distribution characteristics and expansion types of each potential group. In bacterial disease potential groups, lesion clusters are predominantly of the vein-spreading type, and their disease-specific evolutionary trend is marked as extending along the main vein, indicating that these lesions tend to spread along the leaf vein vascular system towards both ends of the leaf. In fungal disease potential groups, lesion clusters are predominantly of the non-directional and general expansion types, and their disease-specific evolutionary trend is marked as scattered in the interveinal region, indicating that these lesions tend to spread in the mesophyll tissue between adjacent leaf veins.

[0070] S107. Assess the degree of agreement between the disease-specific evolution trend and the pre-defined typical morphology of bacterial diseases. When the degree of agreement is high, classify the disease into the bacterial disease group to obtain preliminary disease type classification results.

[0071] Based on the determination of disease-specific evolution trends, standard ranges for the ratio of long to short axes of lesions, the angle between the long axis of the lesion and the main vein, and the proportion of vein-extending types are extracted from the pre-defined typical morphological descriptions of bacterial diseases. For lesion clusters within each potential group of bacterial diseases, the average ratio of long to short axes and the average angle between the main veins of all lesion samples within the cluster are statistically analyzed from the expanded morphological quantification data. Combined with the proportion of vein-extending types in the cluster, the morphological feature data to be evaluated is obtained. The morphological feature data to be evaluated is compared item by item with the standard ranges to determine whether the average ratio of long to short axes, the average angle between the main veins, and the proportion of vein-extending types fall within the corresponding standard ranges. The number of feature items falling within the standard ranges is counted, and the ratio of the number of feature items falling within the ranges to the total number of feature items is calculated as the degree of fit. If the degree of fit is greater than a pre-defined degree of fit threshold, the degree of fit of the lesion cluster is determined to be high. For lesion clusters with a high degree of agreement, they are marked as bacterial disease groups. For lesion clusters with a low degree of agreement, their potential bacterial disease group markers are retained. At the same time, the group markers of all lesion clusters within the potential fungal disease groups are retained. The group classification status of all lesion clusters is summarized to obtain preliminary disease type classification results.

[0072] After the initial classification of potential bacterial and fungal disease groups, further verification of lesion clusters within the bacterial disease groups is required. This is done by comparing these clusters with pre-defined typical morphological descriptions of bacterial diseases to identify lesion clusters whose morphological characteristics highly match those of bacterial diseases. The pre-defined typical morphological descriptions of bacterial diseases are reference standards established based on pathological studies of cassava bacterial black rot, and include numerical ranges across three dimensions.

[0073] In one embodiment, the standard range for the ratio of the long axis to the short axis of lesions in the typical morphological description of bacterial diseases reflects the elongation of the lesions. Lesions formed when bacterial diseases rapidly expand along the vascular bundles of leaf veins typically exhibit a distinctly elongated shape, with a ratio value in the higher range. The standard range for the angle between the long axis of the lesion and the midrib reflects the consistency between the direction of lesion expansion and the direction of the midrib. The long axis of typical bacterial disease lesions is almost parallel to the midrib, with an angle value in the lower range. The standard range for the proportion of lesions extending along the veins reflects the proportion of lesions extending along the veins within a cluster. A higher proportion of lesions in a bacterial disease cluster should exhibit vein-extending characteristics.

[0074] It should be noted that the statistical process for evaluating morphological characteristic data is performed on lesion clusters within each potential group of bacterial diseases. The major and minor axis ratios of all lesion samples within the cluster are read from the extended morphological quantification data. These values ​​are summed and divided by the number of lesion samples to obtain the average major and minor axis ratio. The average midrib angle is calculated using the same method. The percentage of vein extension types has already been calculated in the previous steps and can be directly read and used.

[0075] Specifically, the degree of agreement is calculated through item-by-item comparison. The average aspect ratio is compared to the standard range for lesion aspect ratios to determine if it falls between the lower and upper limits of the standard range. The average midrib angle is compared to the standard range for the angle between the lesion's long axis and the midrib. The percentage of lesion-associated extension types is compared to the standard range for the percentage of lesion-associated extension types. The number of features falling within the corresponding standard range is counted. Since there are three features in total, the degree of agreement is calculated by dividing the number of features falling within the range by three. When the degree of agreement is greater than a preset threshold, the lesion cluster is considered to have a high degree of agreement with the typical morphology of bacterial diseases.

[0076] For example, the preliminary disease type classification results are summarized, and all lesion clusters are organized according to their group affiliation. Lesion clusters with a high degree of conformity are marked as bacterial disease groups, indicating that the lesions within the cluster have typical morphological characteristics of bacterial diseases. Lesion clusters with a low degree of conformity retain the potential bacterial disease group label, indicating that although the cluster conforms to the characteristics of bacterial diseases in terms of the proportion of vein extension types, the overall conformity of morphological characteristics still needs further confirmation. Lesion clusters within the potential fungal disease groups retain their original group labels.

[0077] S108. Extract unclassified groups from the preliminary disease type classification results, assess their similarity to typical morphological characteristics of fungal diseases, and output the results of distinguishing between bacterial and fungal disease types in the scenario of multiple diseases coexisting.

[0078] From the preliminary disease type classification results, lesion clusters retaining potential bacterial disease group markers and lesion clusters retaining potential fungal disease group markers are extracted as unclassified groups. For lesion clusters within the potential fungal disease group, the sum of the number of lesion samples carrying non-directional expansion type markers and ordinary expansion type markers within the cluster is counted. The ratio of this sum to the total number of lesion samples within the cluster is calculated as the non-vein expansion ratio. If the non-vein expansion ratio is greater than a preset fungal ratio threshold, the lesion cluster is marked as a fungal disease group; otherwise, the original potential group marker is retained. The bacterial disease groups, fungal disease groups, lesion clusters retaining potential bacterial disease group markers, and lesion clusters retaining potential fungal disease group markers are summarized, and the spatial distribution and final group assignment status of each lesion cluster are recorded. The results of distinguishing between bacterial and fungal disease types in a multi-disease coexistence scenario are output.

[0079] After confirming the bacterial disease groups, the preliminary disease classification results still contained clusters of lesions retaining potential group markers. These unclassified groups included clusters of lesions from potential bacterial disease groups with a low degree of conformity, as well as clusters of lesions from all potential fungal disease groups. Further evaluation was conducted on the lesion clusters within the potential fungal disease groups to determine whether they conformed to the morphological characteristics of fungal diseases.

[0080] In one embodiment, the calculation of the non-vein spread percentage is performed on lesion clusters within each potential fungal disease group. The number of lesion samples carrying non-directional spread type markers within the cluster is counted, and the number of lesion samples carrying ordinary spread type markers is added to this count. This sum is then divided by the total number of lesion samples within the cluster to obtain the non-vein spread percentage. When this percentage exceeds a preset fungal percentage threshold, it indicates that most lesions within the cluster exhibit non-vein spread morphological characteristics, consistent with the biological characteristics of fungal diseases spreading radially or irregularly in leaf tissues. The lesion cluster is then labeled as a fungal disease group.

[0081] For example, the differentiation results of bacterial and fungal disease types in a multi-disease coexistence scenario include four types of lesion clusters: bacterial disease group, fungal disease group, lesion clusters retaining potential bacterial disease group markers, and lesion clusters retaining potential fungal disease group markers. This differentiation result records the spatial distribution of each lesion cluster on the leaf and its final group assignment, thus achieving the differentiation of bacterial and fungal diseases on leaves with multiple coexisting diseases.

[0082] If the technical solution of this application involves the processing of personal information, the relevant products have established a sound user authorization mechanism: before collecting, using, or sharing personal information, the obligation to inform is fulfilled in accordance with the law, and the individual's voluntary and explicit consent is obtained; if sensitive personal information is involved, the user's separate and explicit consent is further obtained. Specific measures include, but are not limited to: setting up prominent prompts in the information collection area, or clearly displaying the processing rules (including the processor, purpose, method, information type, etc.) through electronic interfaces such as pop-ups, checkboxes, and active submissions, to ensure that users voluntarily authorize based on their knowledge. All personal information processing activities strictly comply with national laws and regulations, especially the relevant provisions of the "Personal Information Protection Law of the People's Republic of China," to effectively safeguard the legitimate rights and interests of personal information subjects.

[0083] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A cassava leaf trait and disease synergistic analysis method, characterized by, The method comprises: The method comprises: According to the initial data set of lesion morphology, the vein network structure of the leaf is obtained by extracting the vein skeleton, the angle value between the long axis of the lesion and the adjacent main vein and the local inter-vein distance value are identified, and the extended morphological quantitative data is determined; Group the extended morphological quantitative data, combine the vein density distribution data, analyze the density of the local inter-vein distance value, and divide the high-density vein network area and the low-density vein network area; Calculate the angle distribution dispersion of the long axis direction of the lesion sample in the high-density vein network area and the trend of multiple adjacent secondary leaf veins, determine the dispersion threshold value based on the number of leaf vein branches and the angle from the lesion centroid to the leaf vein network node, and determine the non-directional expansion mode when the dispersion is greater than the dispersion threshold value, and determine the real vein expansion feature based on the average vertical distance from the lesion edge to the nearest main vein when the dispersion is less than the dispersion threshold value and the angle between the long axis and the main vein is less than the preset angle threshold value; According to the screening result of the real vein expansion feature, the proportion of the significant vein extension type lesion is analyzed, and the preliminary disease type classification result is determined.

2. The method for synergistically analyzing a cassava leaf trait and a disease according to claim 1, characterized in that, The method comprises: The method comprises: According to the initial data set of lesion morphology, the vein network structure of the leaf is obtained by extracting the vein skeleton, the angle value between the long axis of the lesion and the adjacent main vein and the local inter-vein distance value are identified, and the extended morphological quantitative data is determined; According to the initial data set of lesion morphology, the vein network structure of the leaf is obtained by extracting the vein skeleton, the angle value between the long axis of the lesion and the adjacent main vein and the local inter-vein distance value are identified, and the extended morphological quantitative data is determined; 3. The method according to claim 1, wherein the method is characterized by, For each closed contour in the set of lesion area boundary contours, fit the minimum area circumscribed ellipse, calculate the long-short axis ratio value and the long axis direction angle value, pair the long-short axis ratio value and the long axis direction angle value, and store them to obtain the initial data set of lesion morphology. The method comprises: According to the initial data set of lesion morphology, the vein network structure of the leaf is obtained by extracting the vein skeleton, the angle value between the long axis of the lesion and the adjacent main vein and the local inter-vein distance value are identified, and the extended morphological quantitative data is determined; According to the initial data set of lesion morphology, the vein network structure of the leaf is obtained by extracting the vein skeleton, the angle value between the long axis of the lesion and the adjacent main vein and the local inter-vein distance value are identified, and the extended morphological quantitative data is determined; For each lesion sample in the initial data set of lesion morphology, locate the contour centroid coordinates, measure the angle value between the long axis of the lesion and the adjacent main vein and the local inter-vein distance value; The long-short axis ratio value, the long axis direction angle value, and the long-short axis ratio value, the long axis direction angle value of the corresponding lesion sample in the initial data set of lesion morphology are associated to determine the extended morphological quantitative data.

4. The cassava leaf trait and disease synergistic analysis method according to claim 1, characterized in that, The grouping of the extended morphological quantitative data, in combination with the veinlet density distribution data, analyzes the density of the local inter-vein distance values, and divides the high-density vein network area and the low-density vein network area, including: According to the local inter-vein distance value of each lesion sample in the extended morphological quantitative data, the K-means clustering algorithm is used for grouping to obtain two lesion sample clustering clusters; The average value of the local inter-vein distance value in the two clustering clusters is compared to divide the high-density vein network area and the low-density vein network area.

5. The cassava leaf trait and disease synergistic analysis method according to claim 1, characterized in that, The angle distribution dispersion of the long axis direction of the lesion sample in the high-density vein network area and the trend of multiple adjacent secondary leaf veins is calculated, and the leaf vein branch number and angle of the lesion centroid to the leaf vein network node are determined to determine the dispersion threshold value. When the dispersion is greater than the dispersion threshold value, it is determined as a non-directional expansion mode. When the dispersion is less than the dispersion threshold value and the long axis and the main vein have an angle less than a preset angle threshold value, the average vertical distance of the lesion edge to the nearest main vein is determined to determine the real along-vein expansion feature, including: For the lesion sample carrying the high-density vein network area attribution identifier, the secondary leaf vein branch within a preset radius range around the lesion contour centroid is extracted, the angle between the long axis direction of the lesion and the trend of each secondary leaf vein is measured, and the standard deviation of the angle value sequence is calculated as the angle distribution dispersion; According to the lesion contour centroid coordinates, the nearest leaf vein bifurcation node is located, the number of leaf vein branches extending from the bifurcation node is counted, and the corresponding dispersion threshold value is determined according to the number of leaf vein branches; When the angle distribution dispersion is greater than the dispersion threshold value, the lesion sample is marked as a non-directional expansion mode; when the angle distribution dispersion is less than or equal to the dispersion threshold value and the angle between the long axis of the lesion and the adjacent main vein is less than a preset angle threshold value, multiple points are sampled along the lesion contour edge, the average vertical distance of each point to the nearest main vein is calculated, and the real along-vein expansion feature identifier is determined and added according to the average vertical distance.

6. The cassava leaf trait and disease synergistic analysis method according to claim 1, characterized in that, According to the screening result of the real along-vein expansion feature, the proportion of the significantly along-vein extension type lesion is analyzed, and the preliminary disease type division result is determined, including: According to the screening result of the real along-vein expansion feature, when the contour major and minor axis ratio of the lesion sample in the low-density vein network area is greater than a threshold value and the long axis and the main vein have an angle less than a preset angle, it is marked as a significantly along-vein extension type, and the determination results of the high-density vein network area and the low-density vein network area are combined to generate an expansion stage anisotropic distribution map; According to the expansion stage anisotropic distribution map, based on the proportion of the significantly along-vein extension type lesion in each cluster of the multiple disease coexistence lesions, the potential groups of bacterial and fungal diseases are divided, and the disease-specific evolution trend is determined; The consistency degree of the disease-specific evolution trend and the preset typical morphology of the bacterial disease is evaluated, and when the consistency degree is at a high level, it is classified into the bacterial disease group to obtain the preliminary disease type division result.

7. The cassava leaf trait and disease synergistic analysis method according to claim 6, characterized in that, According to the screening result of the real along-vein expansion feature, when the contour major and minor axis ratio of the lesion sample in the low-density vein network area is greater than a threshold value and the long axis and the main vein have an angle less than a preset angle, it is marked as a significantly along-vein extension type, and the determination results of the high-density vein network area and the low-density vein network area are combined to generate an expansion stage anisotropic distribution map, including: According to the real vein expansion feature identifier, the lesion sample carrying the real vein expansion feature identifier is extracted from the high-density vein network area and marked as a vein expansion type, and a non-directional expansion mode mark is retained; For the lesion sample in the low-density vein network area, according to the long-short axis ratio value and the main vein angle value, it is marked as a significant vein extension type or a general expansion type; The lesion expansion type records of the high-density vein network area and the low-density vein network area are combined, the type marks are mapped based on the contour centroid coordinates, and an expansion stage anisotropic distribution map is generated.

8. The cassava leaf trait and disease synergistic analysis method according to claim 6, characterized in that, Based on the expansion stage anisotropic distribution map, the potential group of bacterial and fungal diseases is divided based on the proportion of the significant vein extension type lesion in each cluster of the multi-disease coexistence lesion, and the disease-specific evolution trend is determined, including: According to the lesion contour centroid coordinates in the expansion stage anisotropic distribution map, a hierarchical clustering algorithm is used to group the lesion samples, obtaining a plurality of lesion clusters; For each lesion cluster, the proportion of the significant vein extension type lesion is calculated, and when the proportion value is greater than a preset proportion threshold, the cluster is classified into the potential group of bacterial diseases, otherwise it is classified into the potential group of fungal diseases; According to the division result, the disease-specific evolution trend extending along the main vein direction is marked for the potential group of bacterial diseases, and the disease-specific evolution trend spreading in the inter-vein area is marked for the potential group of fungal diseases, and the disease-specific evolution trend is determined.

9. The cassava leaf trait and disease synergistic analysis method according to claim 6, characterized in that, The degree of coincidence between the evaluation of the disease-specific evolution trend and the preset typical morphology of bacterial diseases is determined, and when the degree of coincidence is at a high level, it is classified into the bacterial disease group to obtain a preliminary disease type division result, including: The long-short axis ratio standard range, the main vein angle standard range and the vein extension type proportion standard range are extracted from the preset typical morphology of bacterial diseases; For each lesion cluster in the potential group of bacterial diseases, the average value of the long-short axis ratio, the average value of the main vein angle and the proportion of the vein extension type are calculated; The statistical results are compared with the standard range one by one, and the proportion of the feature items falling within the standard range is calculated as the degree of coincidence value, and when the degree of coincidence value is greater than a preset coincidence threshold, the cluster is marked as a bacterial disease group, and the preliminary disease type division result is obtained.

10. The cassava leaf trait and disease synergistic analysis method according to claim 1, characterized in that, After determining the preliminary disease type division result, including: The lesion cluster with a potential group label is extracted from the preliminary disease type division result as an unclassified group; For the lesion cluster in the potential group of fungal diseases, the proportion of the non-vein expansion type lesion is calculated, and when the proportion value is greater than a preset fungal proportion threshold, it is marked as a fungal disease group; The bacterial disease group, the fungal disease group and the cluster with the potential group label are summarized, the spatial distribution position and the final attribution state are recorded, and the bacterial and fungal disease type differentiation result is output.

Images