Plant image processing-based classification method and system for stress-resistant plants on high-cold and steep slopes

Abstract

The application provides a high-cold and steep-slope stress-resistant plant classification method and system based on plant image processing, and belongs to the fields of image processing and artificial intelligence. Time-series growth morphology images of a target plant group in a preset growth cycle in a high-cold and steep-slope region are collected, the time-series growth morphology images are subjected to morphology response trajectory analysis processing, a plant morphology response trajectory set is obtained through time-series image frame difference and morphology change vector extraction, stress resistance interactive modeling is performed on the plant morphology response trajectory set, a plant stress resistance interactive atlas is generated through multi-feature synergistic effect strength calculation and stress resistance contribution degree adjustment, and the plant morphology response trajectory set and the plant stress resistance interactive atlas are subjected to progressive prototype adaptive matching with a preset stress resistance prototype trajectory mode library to determine a high-cold and steep-slope stress-resistant plant classification result. Through the application, the influence of feature correlation misjudgment on the classification result can be effectively reduced, and the reliability and robustness of stress-resistant plant classification can be improved.

Description

Technical Field

[0001] This invention relates to the field of image processing, and more specifically, to a method and system for classifying stress-resistant plants on cold and steep slopes based on plant image processing. Background Technology

[0002] In ecological environmental protection and vegetation restoration projects in high-altitude and steep slope areas, accurate classification of stress-resistant plants is a key technical link in optimizing vegetation configuration and enhancing ecosystem stability. Its core lies in identifying species with environmental adaptability through plant morphological indicators, providing a scientific basis for vegetation construction in special habitats. Currently, stress-resistant plant classification methods based on plant image processing typically collect static morphological features such as leaf morphology, stem structure, or root distribution, extracting morphological parameters such as leaf area, stem diameter, and root length. These parameters are then used in conjunction with traditional classification models for feature matching or threshold judgment to determine the plant's stress resistance level. However, these methods have certain limitations in practical applications: on the one hand, the morphological features collected are mostly static parameters of specific growth stages, which are difficult to fully reflect the morphological response process of plants under complex stresses such as seasonal temperature differences, wind erosion, and soil nutrient changes in high-altitude and steep slope environments; on the other hand, the extracted morphological parameters are mostly processed by simple combination or fixed weight fusion, which fails to fully consider the synergistic mechanism between different morphological features in stress resistance performance, making it difficult for the classification model to accurately capture the intrinsic relationship between plant morphology and stress resistance when facing the special habitat of high-altitude and steep slopes, thus affecting the adaptability and reliability of the classification results. Summary of the Invention

[0003] This invention provides a method and system for classifying stress-resistant plants on cold and steep slopes based on plant image processing.

[0004] In a first aspect, embodiments of the present invention provide a method for classifying stress-resistant plants on cold and steep slopes based on plant image processing. The method includes: acquiring temporal growth morphology images of a target plant population in a cold and steep slope area within a preset growth cycle, wherein the temporal growth morphology images are sequences of overall plant morphological changes including multiple consecutive growth stages; performing morphological response trajectory analysis on the temporal growth morphology images, and obtaining a set of plant morphological response trajectories including morphological change trajectories of growth stages, stress response inflection point features, and temporal correlation features through inter-frame difference and morphological change vector extraction of the temporal images; performing stress resistance interaction modeling based on the set of plant morphological response trajectories, and generating a plant stress resistance interaction map by calculating the intensity of multi-feature synergistic effects and adjusting the stress resistance contribution; and performing progressive prototype adaptation matching between the set of plant morphological response trajectories and the plant stress resistance interaction map and a preset stress resistance prototype trajectory pattern library to determine the classification result of stress-resistant plants on cold and steep slopes.

[0005] Secondly, embodiments of the present invention provide a computer system, comprising: a memory storing a computer program; and a processor for loading the computer program to implement the above-described method for classifying stress-resistant plants on cold and steep slopes based on plant image processing.

[0006] This invention provides a method for classifying stress-resistant plants on cold and steep slopes based on plant image processing. By acquiring a sequence of images showing the overall morphological changes of plants across multiple continuous growth stages, it can comprehensively capture the growth process of plants in cold and steep slope environments, providing complete temporal data support for subsequent morphological response analysis. The method analyzes the morphological response trajectory of the temporal growth morphology images, extracting the morphological change trajectory of growth stages, stress response inflection point features, and temporal correlation features through inter-frame difference and morphological change vector extraction. This transforms plant morphological features from static morphological descriptions into evolutionary trajectories, while simultaneously capturing key response nodes in morphological changes and the temporal correlation patterns between features, reflecting the evolutionary patterns of plant morphology within the growth cycle and its response patterns under environmental pressure. Based on the set of plant morphological response trajectories, stress resistance interaction modeling is performed. By calculating the intensity of multi-feature synergistic effects and adjusting the stress resistance contribution, a plant stress resistance interaction map is generated. This quantifies the stress resistance synergistic mechanism between different morphological features, realizes the allocation of feature contributions, and clearly presents the stress resistance correlation between morphological features. By progressively adapting and matching the set of plant morphological response trajectories and the stress resistance interaction map with the pre-set stress resistance prototype trajectory pattern library, and combining trajectory trends and feature collaboration mechanisms for deep matching, we can achieve multi-dimensional classification from morphological evolution process to stress resistance mechanism, improve the adaptability and accuracy of classification results to special habitats of high-altitude and steep slopes, and effectively reduce the impact of misjudgment of inter-feature associations on classification results through interactive modeling and progressive matching process, thereby improving the reliability and robustness of stress-resistant plant classification. Attached Figure Description

[0007] Figure 1 This is a flowchart of a method for classifying stress-resistant plants on cold and steep slopes based on plant image processing, provided by an embodiment of the present invention.

[0008] Figure 2 This is a schematic diagram of the composition of a computer system provided in an embodiment of the present invention. Detailed Implementation

[0009] Please see Figure 1 , Figure 1 A flowchart of a method for classifying stress-resistant plants on cold and steep slopes based on plant image processing is provided for embodiments of the present invention. This method can be executed by a computer system and may include the following steps: Step S100: Collect time-series growth morphology images of the target plant population in the high-altitude and steep slope area within a preset growth cycle. The time-series growth morphology images are sequences of overall plant morphological changes that include multiple consecutive growth stages.

[0010] Temporal growth morphology images refer to a series of images continuously captured from a target plant population in a high-altitude, steep slope region within a predetermined growth cycle. These images record the overall morphological changes of the plant from one growth stage to the next. The predetermined growth cycle is a time period determined in advance based on the growth characteristics of the target plant. For example, for some annual plants, the predetermined growth cycle may be the entire process from seed germination to seed maturity; for perennial plants, it may be a growing season. The sequence of images showing the overall morphological changes of the plant is arranged chronologically, reflecting the plant's appearance at different growth stages, such as plant height, the number and shape of leaves, and the direction of branch extension.

[0011] When acquiring time-series growth morphology images, a high-definition camera can be used, and the shooting time interval can be set, such as shooting once every other day.

[0012] Step S200: Perform morphological response trajectory analysis on the time-series growth morphology images. By extracting the morphological change vectors from the inter-frame differences of the time-series images, a set of plant morphological response trajectories containing morphological change trajectories of growth stages, inflection point features of pressure response, and temporal correlation features is obtained.

[0013] Morphological response trajectory analysis involves in-depth analysis of time-series growth morphology images to extract feature information reflecting the morphological changes during plant growth. Inter-frame difference analysis of time-series images calculates the differences between two adjacent frames of time-series growth morphology images, highlighting morphological changes between adjacent growth stages. The morphological change vector describes the direction and magnitude of plant morphological changes; extracting this vector allows for a more precise representation of these changes.

[0014] The morphological change trajectory during growth stages is the path of morphological changes in plants at different growth stages, reflecting the dynamic process of plant growth. The stress response inflection point characteristic refers to the point in the plant's growth process where external stress (such as low temperature or drought) causes a significant turning point in morphological changes. The timestamps and morphological change vectors corresponding to these points can serve as important criteria for identifying plant stress resistance. Temporal correlation characteristics describe the degree of correlation between morphological change trajectories of adjacent growth stages, reflecting the continuity and stability of plant growth.

[0015] As one implementation method, step S200 can be specifically implemented as the following steps S210~S260: Step S210: Perform temporal alignment processing on the temporal growth morphology images. Based on the image acquisition timestamps, calibrate the image frames of multiple growth stages into a temporal image sequence with equal time intervals to generate a temporally aligned image sequence.

[0016] Temporal alignment is performed to eliminate potential time errors during image acquisition, arranging image frames from multiple growth stages at equal time intervals. The image acquisition timestamp is the capture time corresponding to each image frame; this timestamp determines the temporal order and time intervals between image frames. A temporally aligned image sequence refers to an image sequence where adjacent image frames have equal time intervals, facilitating subsequent analysis and processing.

[0017] During time-series alignment, the timestamp of each image frame is first read. Then, the image frames are filtered and interpolated according to a preset time interval. For example, if the preset time interval is one day, but the time interval for some image frames may be two or three days, a linear interpolation method can be used to generate virtual image frames at the missing time points, ensuring that the time interval for all image frames is one day. Specifically, for two adjacent time points with actual images, the content of the image frame at the missing time point is calculated linearly based on their corresponding image frame information.

[0018] Step S220: Perform inter-frame difference processing on the temporally aligned image sequence, calculate the pixel grayscale difference between adjacent image frames, extract the inter-frame morphological change region through threshold segmentation, and generate a binarized image sequence of the morphological change region.

[0019] Inter-frame differential processing highlights morphological changes in plants between adjacent growth stages by calculating the pixel grayscale difference between two temporally aligned frames. The pixel grayscale difference refers to the difference in grayscale values ​​of corresponding pixels in adjacent image frames, reflecting the brightness information of the pixel. Thresholding segmentation is an image segmentation method that divides pixels in an image into two categories based on a preset threshold: pixels above the threshold are considered morphological change regions, and pixels below the threshold are considered background regions. A binarized image sequence of morphological change regions is an image sequence that represents morphological change regions using binary values ​​(such as 0 and 1), where 1 represents a morphological change region and 0 represents a background region.

[0020] During inter-frame difference processing, each pixel in adjacent temporally aligned images is traversed, and their grayscale difference is calculated. For example, for the pixel at position (x,y) in images I1 and I2, the grayscale difference is |I1(x,y)-I2(x,y)|. Then, a suitable threshold is determined experimentally or empirically, and each pixel in the grayscale difference image is compared with this threshold. If the grayscale difference of a pixel is greater than the threshold, the pixel is marked as 1, indicating that it belongs to the morphological change region; if it is less than the threshold, it is marked as 0, indicating that it belongs to the background region. For the above calibrated temporally aligned image sequence, inter-frame difference processing and threshold segmentation are performed according to this method to generate a binary image sequence of morphological change regions.

[0021] Step S230: Perform morphological dilation and erosion processing on each binarized image in the binarized image sequence of morphological change regions to eliminate noise interference and connect the broken change regions to generate a denoised morphological change region image.

[0022] Morphological dilation and erosion are used to process binarized images to improve image quality. Morphological dilation expands the foreground region (i.e., the region of morphological change) outward in a binarized image, which can connect broken areas of change; morphological erosion shrinks the foreground region inward, which can eliminate noise interference. Morphological dilation is first performed on each binarized image in the sequence of binarized images of the morphological change region, followed by erosion, ultimately generating a denoised image of the morphological change region.

[0023] Step S240: Extract the centroid coordinate sequence of the morphological change region from the denoised morphological change region image, calculate the morphological change vector based on the change of centroid coordinates of continuous timestamps, and arrange the morphological change vectors in chronological order to form the morphological change trajectory of the growth stage.

[0024] Centroid coordinates refer to the coordinates of the geometric center of the region undergoing morphological change. By extracting the centroid coordinate sequence of the region, the positional changes of the region at different times can be described. The change in centroid coordinates at consecutive timestamps refers to the difference in centroid coordinates at adjacent time points. Based on this difference, the morphological change vector can be calculated. The direction of the morphological change vector indicates the direction of the morphological change, and the magnitude of the vector indicates the magnitude of the morphological change.

[0025] When extracting the centroid coordinate sequence, the boundary of each morphological change region in the denoised morphological change region image is first determined. Boundary tracking algorithms, such as chain code tracking algorithms, can be used. Starting from a boundary point, points on the boundary are tracked sequentially according to certain rules (such as clockwise or counterclockwise) until the starting point is reached, thus determining the boundary of the entire morphological change region. Then, the centroid coordinates are calculated based on the coordinates of the boundary points. For a morphological change region composed of multiple pixels, its centroid coordinates (x...) are... c,y c The formula for calculating x is: c =(∑x i ) / n, y c =(∑y i ) / n, where x i and y i Here, n represents the coordinates of each pixel on the boundary, and n is the number of boundary points. Based on consecutive timestamps, the difference between adjacent centroid coordinates is calculated to obtain the morphological change vector. For example, if the centroid coordinates at time point t1 are (x1, y1) and at time point t2 are (x2, y2), then the morphological change vector is (x2-x1, y2-y1). Arranging these morphological change vectors in chronological order forms the morphological change trajectory of the growth stage.

[0026] Step S250: Perform inflection point detection processing on the morphological change trajectory during the growth stage. Identify the extreme points in the trajectory by calculating the rate of change of the second derivative of the trajectory curve. Use the timestamp and morphological change vector corresponding to the extreme points as the inflection point features of the pressure response.

[0027] Inflection point detection aims to identify points in the morphological change trajectory during the growth stage where significant inflection points occur. These points may be related to external stress on the plant. The rate of change of the second derivative of the trajectory curve refers to the change in the second derivative of the trajectory curve. By calculating the rate of change of the second derivative, extreme points in the trajectory can be identified more accurately. Extreme points are the maximum or minimum points in the trajectory curve. The timestamps and morphological change vectors corresponding to these points can be used as inflection point features of the stress response.

[0028] As one implementation method, step S250 can be specifically implemented as the following steps S251~S256: Step S251: Represent the morphological change trajectory during the growth stage as a trajectory function with timestamp as the independent variable, and the dependent variables of the trajectory function are the direction component and the magnitude component of the morphological change vector.

[0029] The trajectory function represents the morphological changes during the growth stages in functional form for subsequent calculations and analysis. The timestamp serves as the independent variable, indicating the order of time, while the directional and modulus components of the morphological change vector serve as the dependent variables, reflecting the changes in the direction and magnitude of plant morphological changes over time.

[0030] When representing the morphological change trajectory during the growth stage as a trajectory function, curve fitting can be used. For example, polynomial fitting can be used for the directional component and the magnitude component of the morphological change vector, respectively. Assuming the directional component of the morphological change vector has values ​​d1, d2, ..., dn at different time points t1, t2, ..., tn, a polynomial function f(t) = a0 + a1t + a2t can be used.2 The data points are fitted using the formula +...+amtm. The coefficients a0, a1,...,am of the polynomial are solved using the least squares method to minimize the error between the fitted function f(t) and the actual data points. The same method is used to fit the modulus component. This yields a trajectory function with the timestamp as the independent variable and the direction component and modulus component of the morphological change vector as dependent variables.

[0031] Step S252: Perform second-order derivative operations on the directional component and the magnitude component of the trajectory function to obtain the sequence of second-order derivatives of the directional component and the sequence of second-order derivatives of the magnitude component.

[0032] The second-order derivative operation involves differentiating the trajectory function twice, yielding a sequence of second-order derivatives. The directional second-order derivative sequence is composed of the second-order derivatives of the direction component of the trajectory function, while the modulus second-order derivative sequence is composed of the second-order derivatives of the modulus component. These second-order derivative sequences reflect the changes in the rate of change of the trajectory function and help identify extreme points in the trajectory. By performing second-order derivatives on the polynomial functions of the direction and modulus components of the trajectory function respectively, we obtain the directional and modulus second-order derivative sequences.

[0033] Step S253: Calculate the absolute value of the difference between adjacent elements in the second derivative sequence of the direction. When the absolute value of the difference exceeds the preset direction change threshold, mark the corresponding timestamp as a candidate inflection point of the direction.

[0034] The direction change threshold is a pre-set value based on actual needs or experience. It is used to determine whether the difference between adjacent elements in the second derivative sequence of the direction is large enough to identify a candidate inflection point of the direction. The absolute value of the difference refers to the absolute value of the difference between adjacent elements of the second derivative of the direction. When this absolute value exceeds the direction change threshold, it indicates that the change in the second derivative of the direction is relatively significant, and the corresponding timestamp may be a candidate inflection point of the direction.

[0035] When calculating the absolute value of the difference, the sequence of second derivatives in the direction is traversed, and the difference between adjacent elements is calculated in turn, and the absolute value is taken. For example, for the sequence of second derivatives in the direction d1, d2, ..., dn, |d2-d1|, |d3-d2|, ..., |dn-dn-1| are calculated. Then, these absolute values ​​of difference are compared with a preset direction change threshold. When the absolute value of the difference exceeds the threshold, the corresponding timestamp is marked as a candidate inflection point of the direction.

[0036] Step S254: Calculate the absolute value of the difference between adjacent elements in the second derivative sequence of the modulus length. When the absolute value of the difference exceeds the preset modulus length change threshold, mark the corresponding timestamp as a candidate inflection point of the modulus length.

[0037] The modulus change threshold is a pre-set value based on actual needs or experience. It is used to determine whether the difference between adjacent elements in the modulus second derivative sequence is large enough to identify a candidate inflection point for the modulus. The absolute value of the difference refers to the absolute value of the difference between adjacent modulus second derivative elements. When this absolute value exceeds the modulus change threshold, it indicates that the change in the modulus second derivative is significant, and the corresponding timestamp may be a candidate inflection point for the modulus.

[0038] When calculating the absolute value of the difference, the method is the same as that used to calculate the absolute value of the difference in the sequence of second derivatives of the direction. The sequence of second derivatives of the modulus length is traversed, and the difference between adjacent elements is calculated in turn, and the absolute value is taken. Then, these absolute values ​​of difference are compared with a preset threshold for modulus length change. When the absolute value of the difference exceeds the threshold, the corresponding timestamp is marked as a candidate inflection point for modulus length.

[0039] Step S255: Perform timestamp intersection calculation on the candidate inflection points of direction and candidate inflection points of magnitude, and determine the timestamp that simultaneously satisfies the thresholds of direction and magnitude change as the pressure response inflection point.

[0040] Timestamp intersection operation identifies points with identical timestamps among candidate inflection points for direction and candidate inflection points for modulus. These points simultaneously satisfy the thresholds for both direction and modulus changes, indicating that at these time points, the plant's morphological changes show significant inflections in both direction and modulus. Therefore, these timestamps can be identified as stress response inflection points. During the timestamp intersection operation, the lists of timestamps for candidate inflection points for direction and candidate inflection points for modulus are sorted, and then the timestamps in the two lists are compared sequentially. If identical timestamps are found, they are identified as stress response inflection points. For example, a sorting algorithm (such as quicksort) can be used to sort the two timestamp lists, and then a two-pointer method can be used to traverse the two lists to find identical timestamps.

[0041] Step S256: Extract the morphological change vector corresponding to the pressure response inflection point, including the angle value of the direction component and the length ratio of the magnitude component, and combine them to form the pressure response inflection point feature.

[0042] The morphological change vector corresponding to the inflection point of the pressure response refers to the morphological change vector at the time point corresponding to the inflection point of the pressure response. Extracting the angle value of the directional component and the length ratio of the modulus component of this vector can more comprehensively describe the morphological changes of the plant at the inflection point of the pressure response. Combining these angle values ​​and length ratios forms the pressure response inflection point feature. When extracting the morphological change vector corresponding to the pressure response inflection point, the corresponding morphological change vector is found from the morphological change trajectory of the growth stage based on the timestamp of the pressure response inflection point. Then, the angle value of the directional component and the length ratio of the modulus component of this vector are calculated. For vector (x, y), the angle value θ of its directional component can be calculated using trigonometric functions: θ = arctan(y / x) (quadrant adjustment is required depending on the signs of x and y). The length ratio of the modulus component can be calculated by the ratio of the modulus of the vector to a certain reference modulus, which can be the average of the moduli of all morphological change vectors. Combining these values ​​forms the pressure response inflection point feature. For a given pressure response inflection point, the corresponding morphological change vector is found from the morphological change trajectory of the growth stage based on its timestamp, and the angle value of its directional component and the length ratio of its modulus component are calculated and combined to form the pressure response inflection point feature.

[0043] Step S260: Calculate the cosine similarity of morphological change trajectories of adjacent growth stages, generate temporal correlation features between trajectories, unify the dimensions of morphological change trajectories of growth stages, inflection point features of pressure response, and temporal correlation features, and generate a set of plant morphological response trajectories.

[0044] Cosine similarity is an index used to measure the degree of similarity between two vectors. By calculating the cosine similarity of morphological change trajectories at adjacent growth stages, the temporal correlation characteristics between the trajectories can be obtained, reflecting the degree of similarity between morphological change trajectories at adjacent growth stages. Dimensional unification refers to processing the morphological change trajectories at growth stages, the inflection point characteristics of pressure response, and the temporal correlation characteristics to make them have the same dimension for subsequent analysis and processing. In performing dimensional unification, a feature scaling method is used for the morphological change trajectories at growth stages, the inflection point characteristics of pressure response, and the temporal correlation characteristics. For example, for each feature, its maximum and minimum values ​​are calculated, and then each feature value is scaled to a fixed range (e.g., [0,1]). The specific scaling formula is: x'=(x-min) / (max-min), where x is the original feature value, min and max are the minimum and maximum values ​​of the feature, and x' is the scaled feature value. These processed features are combined to generate a set of plant morphological response trajectories.

[0045] Step S300: Based on the set of plant morphological response trajectories, perform stress resistance interaction modeling, and generate plant stress resistance interaction map by calculating the intensity of multi-feature synergistic effects and adjusting the stress resistance contribution.

[0046] Stress resistance interaction modeling analyzes the set of plant morphological response trajectories to reveal the interactions between different features in plant stress resistance. Multi-feature synergistic effect calculation involves comprehensively considering multiple features in the plant morphological response trajectory set and calculating the synergistic effect strength between them to understand the combined effect of these features in stress resistance. Stress resistance contribution adjustment adjusts the contribution of each feature in the stress resistance process based on the multi-feature synergistic effect strength to more accurately assess the importance of each feature. Plant stress resistance interaction maps are a visual representation that graphically displays the interaction relationships and contributions between various features in plant stress resistance.

[0047] As one implementation method, step S300 can be specifically implemented as the following steps S310~S350: Step S310: Extract the morphological change trajectory of growth stage, the inflection point feature of pressure response, and the temporal correlation feature from the plant morphological response trajectory set, and construct a three-dimensional feature interaction matrix. The row dimension of the matrix is ​​the growth stage, the column dimension is the feature type, and the depth dimension is the feature value.

[0048] A three-dimensional feature interaction matrix is ​​a matrix used to represent the relationships between features in a set of plant morphological response trajectories. The row dimension represents the growth stage, the column dimension represents the feature type (such as morphological change trajectory of growth stage, inflection point feature of pressure response, temporal correlation feature, etc.), and the depth dimension represents the specific value of the feature. By constructing a three-dimensional feature interaction matrix, it is easier to analyze and process each feature.

[0049] When constructing the 3D feature interaction matrix, the number of growth stages, the number of feature types, and the numerical range of each feature are first determined. Then, a 3D array is created, and the data of the growth stage morphological change trajectory, pressure response inflection point features, and time-series correlation features are filled into the array according to the order of growth stage, feature type, and feature value. For example, for growth stage i, feature type j, and its corresponding feature value k, the matrix is ​​filled at the i-th row, j-th column, and k-th depth position.

[0050] Step S320: Perform pressure-sensitive interval identification processing on the three-dimensional feature interaction matrix. By scanning the distribution of pressure response inflection points in the matrix through a sliding window, determine the continuous growth stage where the feature interaction intensity exceeds a preset threshold and mark it as a pressure-sensitive interval.

[0051] The stress-sensitive region identification process aims to pinpoint time periods during plant growth when they are highly sensitive to stress. These periods exhibit high feature interaction intensity, potentially closely related to the plant's stress resistance. A fixed-size window is slid across a three-dimensional feature interaction matrix, calculating the distribution of stress response inflection points and feature interaction intensity within the window. A preset threshold, determined based on practical needs or experience, is used to assess whether the feature interaction intensity is sufficiently high. When the feature interaction intensity exceeds this threshold, the corresponding continuous growth stage is marked as a stress-sensitive region.

[0052] As one implementation method, step S320 can be specifically implemented as the following steps S321~S325: Step S321: Set the time sliding window size to the preset number of growth stages, and slide it along the row dimension of the three-dimensional feature interaction matrix with a window step size as one growth stage.

[0053] The time sliding window size refers to the number of growth stages covered by the sliding window. The preset number of growth stages is determined in advance based on experiments or experience. The window step size refers to the number of growth stages the sliding window moves in each step; here, it is set to one growth stage, meaning the window moves forward one growth stage at a time. By sliding the window along the row dimension of the three-dimensional feature interaction matrix, different growth stages can be analyzed.

[0054] When setting the time-sliding window size, the preset number of growth stages is determined based on the plant's growth characteristics and research objectives. For example, for plants with short growth cycles, the window size can be set to 5 growth stages; for plants with longer growth cycles, the window size can be set to 10 growth stages. Then, starting from the first row of the 3D feature interaction matrix, the window slides sequentially along the row dimension of the matrix, with a step size of one growth stage. For example, a loop structure can be used, increasing the starting row of the window by 1 each time, until the window slides to the last row of the matrix.

[0055] Step S322: Calculate the density value of the pressure response inflection point within each window, and represent the density value by the ratio of the number of inflection points to the window size.

[0056] The density value of pressure response inflection points reflects the distribution of pressure response inflection points within a window. By calculating the density value, we can understand the concentration of pressure response inflection points at different growth stages. The number of inflection points refers to the number of pressure response inflection points within the window, and the window size refers to the number of growth stages covered by the time sliding window. The density value of pressure response inflection points is obtained by dividing the number of inflection points by the window size.

[0057] When calculating the density value of pressure response inflection points, the process iterates through each growth stage within the window, counting the number of pressure response inflection points. For example, a counter is used, incrementing by 1 when a pressure response inflection point is encountered within the window. The counter value is then divided by the window size to obtain the density value. For the distribution of pressure response inflection points within the sliding window, a counter is used to count the number of inflection points and calculate the density value.

[0058] Step S323: Extract the interaction covariance matrix of all feature types within the window, and calculate the trace of the covariance matrix as the feature interaction strength index.

[0059] The interaction covariance matrix is ​​a matrix used to measure the relationships between different feature types within a window, reflecting the collaborative changes among the features. The trace of the covariance matrix is ​​the sum of the elements on the main diagonal of the matrix. By calculating the trace of the covariance matrix, a comprehensive feature interaction strength index can be obtained, which reflects the interaction strength between features within the window.

[0060] When extracting the interaction covariance matrix for all feature types within a window, the feature data within the window is first organized according to feature type, forming a two-dimensional matrix. Rows in the matrix represent growth stages, and columns represent feature types. Then, the covariance between any two feature types is calculated using the general formula for covariance. After calculating the covariance matrix, the elements on the main diagonal of the matrix are summed to obtain the trace of the covariance matrix, which serves as the feature interaction strength index. For feature data within the sliding window, the interaction covariance matrix and its trace are calculated using the same method to obtain the feature interaction strength index.

[0061] Step S324: When the inflection point density value of the window exceeds the density threshold and the feature interaction intensity index exceeds the intensity threshold, mark the window as a candidate sensitive region.

[0062] The density threshold and intensity threshold are two pre-set values ​​used to determine whether the inflection point density value and feature interaction intensity index of the window are sufficiently high. When the inflection point density value of the window exceeds the density threshold and the feature interaction intensity index exceeds the intensity threshold, it indicates that the pressure response inflection point distribution within the window is relatively concentrated and the feature interaction intensity is high. Therefore, the window is marked as a candidate sensitive region.

[0063] During the judgment process, the inflection point density value of the window is compared with a density threshold, and the feature interaction strength index is compared with an intensity threshold. If both conditions are met, the window is marked as a candidate sensitive region. For example, this judgment process can be implemented using conditional statements. For the calculated inflection point density value and feature interaction strength index of the window, conditional statements are used to compare them with the density threshold and intensity threshold to mark candidate sensitive regions.

[0064] Step S325: Merge adjacent candidate sensitive intervals to form a pressure response sensitive interval in a continuous growth stage. The start of the interval is the starting stage of the first candidate window, and the end of the interval is the ending stage of the last candidate window.

[0065] Merging processes connect adjacent candidate sensitive regions to form a continuous stress response sensitive region, which more accurately represents the period during which the plant is sensitive to stress during its growth. The start and end points of the interval represent the beginning and end of the growth stage of the stress response sensitive region, respectively.

[0066] During the merging process, the candidate sensitive intervals are first sorted according to their initial growth stages. A sorting algorithm (such as bubble sort) can be used to sort the candidate sensitive intervals. Then, the sorted candidate sensitive intervals are traversed, checking whether adjacent intervals are adjacent. If they are adjacent, they are merged into one interval. Specifically, if the ending stage of one candidate sensitive interval is adjacent to the starting stage of the next candidate sensitive interval, the two intervals are merged, and the starting stage of the merged interval is updated to the starting stage of the first interval, and the ending stage is updated to the ending stage of the second interval.

[0067] Step S330: Calculate the strength of multi-feature synergy within the pressure response sensitive range. Construct a synergy network using the mutual information entropy and partial correlation coefficient between features. Network nodes represent feature types, and edge weights represent the synergy strength.

[0068] The calculation of multi-feature synergy strength is used to assess the degree of synergy between different feature types within the stress response sensitivity range. Mutual information entropy is an indicator used to measure the degree of information sharing between two random variables; by calculating the mutual information entropy between features, the degree of nonlinear correlation between features can be obtained. Partial correlation coefficient is an indicator that measures the degree of linear correlation between two variables, while keeping other variables constant. A synergy network is a graph structure with feature types as nodes and synergy strength as edge weights. By constructing a synergy network, the synergistic relationships between features can be visually displayed.

[0069] As one implementation method, step S330 can be specifically implemented as the following steps S331~S335: Step S331: Within the pressure response sensitive range, extract five feature types from the three-dimensional feature interaction matrix: the direction component of the growth stage morphological change trajectory, the modulus component, the timestamp of the pressure response inflection point feature, the vector value, and the similarity value of the temporal correlation feature.

[0070] Within the stress response sensitivity range, these feature types can more accurately reflect the growth and stress resistance of plants under stress. Extracting data from these feature types from the three-dimensional feature interaction matrix can provide a foundation for subsequent calculations of synergistic effect strength.

[0071] When extracting data for these feature types, corresponding data are selected from the three-dimensional feature interaction matrix based on the start and end growth stages of the pressure response sensitive interval. For example, for the directional component of the morphological change trajectory during the growth stage, the directional component feature values ​​corresponding to each growth stage within the pressure response sensitive interval are extracted from the matrix. The same method is used to extract other feature types.

[0072] Step S332: Calculate the mutual information entropy between any two feature types. The mutual information entropy is used to measure the degree of nonlinear correlation between features. The larger the mutual information entropy value, the stronger the nonlinear correlation.

[0073] Mutual information entropy is an indicator used to measure the degree of information sharing between two random variables. By calculating the mutual information entropy between any two feature types, the degree of non-linear correlation between them can be obtained. The larger the mutual information entropy value, the stronger the non-linear correlation between the two features.

[0074] When calculating mutual information entropy, the data for each feature type is first discretized, dividing the continuous feature values ​​into several intervals, with each interval corresponding to a discrete value. Then, the frequency of discrete values ​​for each feature type and the frequency of combinations of discrete values ​​from two feature types are calculated. According to the formula for calculating mutual information entropy: I(X;Y)=∑∑p(x,y)log(p(x,y) / (p(x)p(y))), where X and Y are two feature types, p(x,y) is the joint probability distribution of X and Y, and p(x) and p(y) are the marginal probability distributions of X and Y.

[0075] Step S333: Calculate the partial correlation coefficient between any two feature types. While keeping other feature types constant, measure the degree of linear association between the two features. The larger the absolute value of the partial correlation coefficient, the stronger the linear association.

[0076] The partial correlation coefficient is an indicator that measures the degree of linear association between two variables, while keeping other variables constant. By calculating the partial correlation coefficient between any two feature types, the degree of linear association between them can be obtained. The larger the absolute value of the partial correlation coefficient, the stronger the linear association between the two features.

[0077] Suppose we want to calculate the partial correlation coefficient between feature types X and Y, while controlling for other feature types. Unchanged. The specific calculation process is as follows:

[0078] First, two multiple linear regression models are established. The first model uses X as the dependent variable, and Y as the dependent variable. As the independent variable, the expression is ,in It is the intercept term. It is the regression coefficient. This is the error term; the second model uses Y as the dependent variable, and X and As the independent variable, the expression is ,in It is the intercept term. It is the regression coefficient. This is the error term. The least squares method is used to estimate the regression coefficients in these two models. The principle of the least squares method is to minimize the sum of squared errors between the observed values ​​and the model predictions. By solving for the partial derivative of the sum of squared errors with respect to the regression coefficients and setting it to zero, the estimated values ​​of the regression coefficients are obtained.

[0079] Then, calculate the simple correlation coefficient. The formula for calculating (where A and B are any two variables) is: Where m is the sample size, A j and B j These are the values ​​of variables A and B in the j-th sample, respectively. and Let A and B be the sample means of variables A and B, respectively. Then, r is calculated using this formula. XY ,rXZ_i} and (i=1,2,…,n). Finally, substitute the calculated simple correlation coefficient into the partial correlation coefficient formula. In the process, X and Y are obtained under control. The partial correlation coefficient.

[0080] For example, in the classification study of stress-resistant plants on high-altitude and steep slopes, if we want to calculate the partial correlation coefficient of the directional component of the morphological change trajectory of the growth stage and the timestamp of the inflection point feature of the pressure response, we need to keep other features (such as the modulus component of the morphological change trajectory of the growth stage, the vector value of the inflection point feature of the pressure response, and the similarity value of the temporal correlation feature) constant.

[0081] Step S334: The mutual information entropy value and the absolute value of the partial correlation coefficient are weighted and fused. The weight allocation is determined according to the proportion of nonlinear characteristics of the feature type to generate the synergistic effect strength value.

[0082] Weighted fusion combines the mutual information entropy value and the absolute value of the partial correlation coefficient to obtain a comprehensive synergistic effect strength value. The weight allocation is determined based on the proportion of nonlinear characteristics of each feature type; that is, the weights of the mutual information entropy value and the absolute value of the partial correlation coefficient are determined according to the degree of nonlinearity of the feature type. Through weighted fusion, the synergistic effect strength between features can be evaluated more comprehensively.

[0083] When determining the proportion of nonlinear characteristics in a feature type, it can be done by analyzing the distribution and variation patterns of the feature. For example, if a feature exhibits a clear nonlinear trend, such as having multiple peaks or abrupt changes, then it can be considered that the nonlinear characteristic of that feature has a high proportion. For feature types with a high proportion of nonlinear characteristics, the weight of the mutual information entropy value is relatively large; for feature types with a high proportion of linear characteristics, the weight of the absolute value of the partial correlation coefficient is relatively large.

[0084] Assuming the mutual information entropy is H, the absolute value of the partial correlation coefficient is |ρ|, and the proportion of nonlinear characteristics is α (0≤α≤1), then the formula for calculating the synergistic effect strength value S is: S=αH+(1-α)|ρ|.

[0085] Step S335: Using five feature types as network nodes and the cooperative effect strength value as the weight of the directed edge between nodes, construct a cooperative effect network containing self-regulating loops. The self-regulating loops represent the lag effect strength of the feature on itself.

[0086] A collaborative interaction network is a graph structure where feature types are nodes and the strength of the collaboration is the weight of the directed edges. A self-regulating cycle is a directed edge pointing from a node to itself, representing the strength of the lagged effect of a feature on itself. By constructing a collaborative interaction network, the collaborative relationships and self-regulating mechanisms between features can be visually demonstrated.

[0087] When constructing the synergistic network, five nodes are first created, representing the directional component, magnitude component, timestamp, vector value, and similarity value of the temporal correlation feature of the morphological change trajectory during the growth stage. Then, based on the calculated synergistic strength value, a directed edge is added between every two nodes, with the edge weight being the synergistic strength value. For each node, a self-regulating loop is also added. The weight of the self-regulating loop can be determined by analyzing the autocorrelation of the feature at different time points. For example, the correlation coefficient of the feature at adjacent time points can be calculated and used as the weight of the self-regulating loop.

[0088] Step S340: Adjust the resilience contribution weights of each feature based on the edge weights of the collaborative network. The higher the strength of the collaborative effect, the greater the increment of the contribution weight of the corresponding feature, and generate a weight adjustment sequence.

[0089] The stress resistance contribution weight is an indicator used to measure the importance of each feature in the plant's stress resistance process. Adjusting the stress resistance contribution weight of each feature based on the edge weight values ​​of the synergistic network is because feature combinations with higher synergistic strength indicate a more significant synergistic effect during stress resistance; therefore, the contribution weight of the corresponding feature should be increased. The weight adjustment sequence is the sequence of adjusted stress resistance contribution weights for each feature.

[0090] When adjusting the resilience contribution weights of each feature, an initial resilience contribution weight is first assigned to each feature. Then, each edge of the collaborative network is traversed, and the contribution weight increment of the corresponding feature is calculated based on the edge's weight value. Assuming the edge weight is w, and the corresponding two features are A and B, with initial resilience contribution weights wA and wB respectively, the contribution weight increment can be expressed as ΔwA = w × k1, ΔwB = w × k2, where k1 and k2 are pre-set coefficients based on the feature's characteristics and importance. The contribution weight increment is added to the initial resilience contribution weight to obtain the adjusted weights. For example, the adjusted weights wA' = wA + ΔwA, wB' = wB + ΔwB. This adjustment is performed on all features sequentially, generating a weight adjustment sequence.

[0091] Step S350: The three-dimensional feature interaction matrix, pressure response sensitive region, synergistic network and weight adjustment sequence are integrated into a graph. The feature interaction direction is represented by a directed graph and the node size represents the contribution weight to generate a plant stress resistance interaction graph.

[0092] Graphical integration integrates three-dimensional feature interaction matrices, stress response sensitive regions, synergistic networks, and weight adjustment sequences to graphically represent the interaction relationships and contributions of various features in plant stress resistance. A directed graph is a graphical structure used to represent the direction of feature interactions, with node size representing contribution weights. This approach provides a more intuitive view of the importance and interaction relationships of each feature.

[0093] When performing graph-based integration, the nodes and edges of the directed graph are first determined. Nodes represent feature types, including the directional component and magnitude component of the morphological change trajectory during the growth stage, the timestamp and vector value of the inflection point feature of the pressure response, and the similarity value of the temporal correlation feature. Edges represent the interaction relationships between features, the direction of the edge indicates the direction of the interaction, and the weight of the edge can be determined based on the edge weight values ​​of the cooperative network.

[0094] The size of a node is determined by the contribution weights of each feature in the weighted sequence. A mapping function can be designed to map the contribution weights to the node size range. For example, the contribution weights can be normalized to the [0,1] interval, and then linearly mapped to the node size range based on the normalized weight values, such as node radii ranging from 5 to 20 pixels.

[0095] The information of the three-dimensional feature interaction matrix and the stress response sensitive region can be added to the directed graph using annotations or color coding. For example, feature interactions within the stress response sensitive region can be represented by edges of different colors. In this way, the information of the three-dimensional feature interaction matrix, the stress response sensitive region, the synergistic network, and the weight adjustment sequence are integrated into the directed graph to generate a plant stress resistance interaction map.

[0096] Step S400: The set of plant morphological response trajectories and the plant stress resistance interaction map are progressively matched with the preset stress resistance prototype trajectory pattern library to determine the classification results of stress-resistant plants on high-altitude and steep slopes.

[0097] Progressive prototype adaptation matching is a step-by-step matching method that compares the set of plant morphological response trajectories and plant stress resistance interaction maps with patterns in a pre-set stress resistance prototype trajectory pattern library to find the best-matching prototype pattern, thereby determining the classification results of stress-resistant plants on cold and steep slopes. The stress resistance prototype trajectory pattern library is a pre-established database containing typical trajectory patterns of different stress resistance levels. By matching with this database, the stress resistance of plants can be assessed and classified.

[0098] As one implementation method, step S400 can be specifically implemented as the following steps S410~S450: Step S410: Obtain a preset anti-stress prototype trajectory pattern library. The anti-stress prototype trajectory pattern library contains typical trajectory patterns with different levels of anti-stress. Each typical trajectory pattern contains a prototype morphological trajectory, a prototype sensitive region, a prototype cooperative network, and a prototype weight sequence.

[0099] The pre-built stress resistance prototype trajectory pattern library is a pre-constructed database containing typical trajectory patterns for different stress resistance levels. These patterns were obtained through analysis and summarization of a large number of plant samples. The prototype morphological trajectory is the trajectory of plant morphological changes within the typical trajectory pattern; the prototype sensitive interval is the time period during which the plant is sensitive to stress; the prototype cooperative network is the cooperative network between features; and the prototype weight sequence is the weight sequence of the stress resistance contribution of each feature. When acquiring the pre-built stress resistance prototype trajectory pattern library, data can be read from a file storing the database or a server. For example, if the database is stored as a file, file read operations can be used to load the data into memory. For each typical trajectory pattern, its prototype morphological trajectory, prototype sensitive interval, prototype cooperative network, and prototype weight sequence can be stored in different data structures for subsequent matching operations.

[0100] Step S420: Extract the morphological change trajectory of the growth stage from the plant morphological response trajectory set, perform multiple rounds of prototype screening with the prototype morphological trajectory, and determine the candidate prototype set through trajectory trend matching and sensitive interval verification.

[0101] Multi-round prototype screening involves repeatedly narrowing down the matching range to find the most suitable prototype pattern. Trajectory trend matching compares the trends of the growth stage morphological change trajectory and the prototype morphological trajectory to determine their similarity. Sensitive interval verification checks whether the pressure response sensitive interval corresponding to the growth stage morphological change trajectory matches the prototype's sensitive interval. The candidate prototype set is the collection of suitable prototype patterns obtained after screening.

[0102] As one implementation method, step S420 can be specifically implemented as the following steps S421~S425: Step S421: Match the trajectory trend of the morphological change trajectory during the growth stage with the prototype morphological trajectory, calculate the consistency of the overall evolution direction of the trajectory through the time warping algorithm, and consider the local morphological fluctuation pattern of the trajectory in the key growth stage. The key growth stage is divided according to the typical growth cycle of plants on cold and steep slopes, including the budding stage, tillering stage, flowering stage and dormancy stage.

[0103] Trajectory trend matching is used to determine whether the morphological change trajectory during the growth stage and the prototype morphological trajectory are consistent in the overall evolutionary direction. Time warping is an algorithm used to compare the similarity of two time series; this algorithm can calculate the consistency of the overall evolutionary direction of the trajectories. Critical growth stages are defined based on the typical growth cycle of plants on alpine and steep slopes. Considering the local morphological fluctuation patterns of the trajectories during these critical growth stages can more accurately determine the similarity of the trajectories.

[0104] When performing trajectory trend matching, time warping algorithms (such as dynamic time warping) are used to calculate the consistency of the overall evolution direction of the growth stage morphological change trajectory and the prototype morphological trajectory. Dynamic time warping finds the optimal alignment path between two trajectories, minimizing the sum of the distances between corresponding points. Specifically, a two-dimensional matrix is ​​constructed, where the rows and columns represent the time points of the growth stage morphological change trajectory and the prototype morphological trajectory, respectively, and the matrix elements represent the distance between the two trajectories at corresponding time points. Then, a dynamic programming algorithm is used to calculate the shortest path from the top left corner of the matrix to the bottom right corner; the length of this path represents the similarity between the two trajectories.

[0105] Simultaneously, the local morphological fluctuation patterns of the trajectories during the germination, tillering, flowering, and dormancy stages are analyzed. For each key growth stage, local features of the morphological change trajectory and the prototype morphological trajectory at that stage are extracted, such as the amplitude and frequency of morphological changes. The similarity of these local features is compared; if the similarity is high, the local morphological fluctuation patterns are considered to match at that key growth stage. For example, for the germination stage, the amplitude of morphological changes of the two trajectories at this stage is compared; if the difference between the two is within a preset threshold range, the local morphological fluctuation patterns at the germination stage are considered to match.

[0106] Step S422: Select multiple candidate prototypes with the highest similarity ranking based on the trajectory trend matching results as the initial candidate set. The size of the initial candidate set is determined according to the total number of prototypes in the resilience prototype trajectory pattern library, so that the initial candidate set covers prototypes of all resilience levels, and each resilience level contains at least one prototype.

[0107] The initial candidate set is a collection of prototype patterns obtained after preliminary screening. Based on the trajectory trend matching results, several candidate prototypes with the highest similarity ranking are selected, which narrows down the matching range. The size of the initial candidate set is determined by the total number of prototypes in the resilience prototype trajectory pattern library. This ensures that the initial candidate set covers prototypes of all resilience levels, with each resilience level containing at least one prototype, thereby improving matching accuracy.

[0108] When selecting the initial candidate set, all prototype patterns are first sorted according to the similarity of their trajectory trend matching. A sorting algorithm (such as bubble sort or quicksort) can be used to sort the prototype patterns. Then, the size of the initial candidate set is determined based on the total number of prototypes in the resilience prototype trajectory pattern library. For example, if the total number of prototypes is N, a proportionality coefficient k is determined based on experience or experimentation, and the size of the initial candidate set is M = k × N. The top M prototypes from the sorted prototype patterns are selected as the initial candidate set. Simultaneously, it is checked whether the initial candidate set contains at least one prototype for each resilience level. If this condition is not met, prototypes for the corresponding resilience level are selected from the remaining prototype patterns and added to the initial candidate set.

[0109] Step S423: Extract the stress response sensitive interval from the plant stress resistance interaction map and perform interval fit verification with the prototype sensitive interval of each prototype in the initial candidate set. This includes calculating the interval start stage difference, interval end stage difference, and interval length ratio. The interval start stage difference is the difference between the start stage of the target sensitive interval and the start stage of the prototype sensitive interval. The interval end stage difference is the difference between the end stage of the target sensitive interval and the end stage of the prototype sensitive interval. The interval length ratio is the ratio of the length of the target sensitive interval to the length of the prototype sensitive interval.

[0110] Interval fit verification is used to check whether the stress response sensitive intervals in the plant stress resistance interaction map match the prototype sensitive intervals of each prototype in the initial candidate set. The interval start stage difference, interval end stage difference, and interval length ratio are indicators used to measure the degree of matching between two intervals. By calculating these indicators, the fit of the intervals can be determined more accurately.

[0111] When performing interval fit verification, the starting and ending stages of the stress response sensitive intervals are first extracted from the plant stress resistance interaction map. Then, for each prototype in the initial candidate set, the starting and ending stages of its prototype sensitive intervals are extracted. The interval starting stage difference, interval ending stage difference, and interval length ratio are calculated. For example, if the starting stage of the target sensitive interval is s1 and the ending stage is e1, and the starting stage of the prototype sensitive interval is s2 and the ending stage is e2, then the interval starting stage difference is |s1-s2|, the interval ending stage difference is |e1-e2|, and the interval length ratio is (e1-s1) / (e2-s2).

[0112] Step S424: When the starting stage difference of the interval is less than the preset stage difference threshold, the ending stage difference of the interval is less than the preset stage difference threshold, and the interval length ratio is within the preset ratio range, the candidate prototype is determined to pass the interval adaptability check; otherwise, it is removed from the initial candidate set.

[0113] The preset stage difference threshold and preset ratio range are values ​​pre-set based on actual needs or experience, used to determine whether the interval start stage difference, interval end stage difference, and interval length ratio are within acceptable ranges. When these conditions are met, it indicates that the prototype sensitive interval of the candidate prototype matches the pressure response sensitive interval in the plant stress resistance interaction map to a high degree, and the candidate prototype is determined to have passed the interval fit verification; otherwise, it is removed from the initial candidate set. During the judgment process, the calculated interval start stage difference, interval end stage difference, and interval length ratio are compared with the preset stage difference threshold and preset ratio range. This judgment process is implemented using conditional statements. For example, if the interval start stage difference is less than the preset stage difference threshold, the interval end stage difference is less than the preset stage difference threshold, and the interval length ratio is within the preset ratio range, then the candidate prototype is considered to have passed the interval fit verification; otherwise, the candidate prototype is removed from the initial candidate set.

[0114] Step S425: Perform temporal correlation feature comparison on the prototypes that pass the interval fit verification, calculate the overall similarity of the trajectory correlation feature sequence between the target plant and the candidate prototypes, and the similarity is represented by the mean absolute error of the correlation feature values ​​of the corresponding stages in the sequence. The mean absolute error is calculated by the arithmetic mean of the absolute values ​​of the differences of the correlation feature values ​​of each stage in the sequence, and finally determine the set of candidate prototypes that meet the multiple screening conditions.

[0115] Temporal correlation feature comparison is used to further screen prototypes that pass the interval fit test. By calculating the overall similarity of the trajectory correlation feature sequences between the target plant and the candidate prototype, their matching degree can be determined more accurately. Mean absolute error (MAO) is an indicator used to measure the similarity between two sequences. It is obtained by calculating the arithmetic mean of the absolute values ​​of the differences in correlation feature values ​​at each stage of the sequence. The smaller the MAO, the more similar the two sequences are.

[0116] When performing temporal correlation feature comparison, the trajectory correlation feature sequence of the target plant is first extracted from the set of plant morphological response trajectories, and the trajectory correlation feature sequence of the candidate prototype is extracted from the prototypes that have passed the interval fit test. Then, the absolute value of the difference between the correlation feature values ​​of corresponding stages in the two sequences is calculated. For example, for the trajectory correlation feature sequence A=(a1,a2,...,an) of the target plant and the trajectory correlation feature sequence B=(b1,b2,...,bn) of the candidate prototype, |a1-b1|,|a2-b2|,...,|an-bn| are calculated. The absolute values ​​of these differences are added together and then divided by the length n of the sequence to obtain the mean absolute error.

[0117] The mean absolute error (MAE) is compared with a preset threshold. If the MAE is less than the threshold, the trajectory correlation feature sequences of the target plant and the candidate prototype are considered to be generally similar. Based on the magnitude of the MAE, candidate prototypes that meet multiple screening criteria are selected to determine the candidate prototype set.

[0118] Step S430: Perform collaborative network structure alignment processing on each candidate prototype in the candidate prototype set, and generate network matching degree by identifying node correspondence and comparing edge weights.

[0119] The collaborative network structure alignment process compares the similarity between the collaborative networks of candidate prototypes and those in the plant stress resistance interaction graph. Node correspondence identification identifies the correspondences between nodes in the two networks, while edge weight comparison compares the weight values ​​of corresponding edges. Network matching degree is an indicator used to measure the similarity between two collaborative networks. By generating network matching degree, candidate prototypes that best match the collaborative networks in the plant stress resistance interaction graph can be selected.

[0120] As one implementation method, step S430 can be specifically implemented as the following steps S431~S437: Step S431: Extract the synergistic network from the plant stress resistance interaction map as the target network, and extract the prototype synergistic network from the candidate prototype as the template network. Both networks contain feature type nodes, directed edges and edge weights. Feature type nodes include directional component nodes and modulus component nodes of the morphological change trajectory of the growth stage, timestamp nodes and vector value nodes of the stress response inflection point feature, and similarity value nodes of the temporal correlation feature.

[0121] The target network and template network are two collaborative networks used for comparison. They contain nodes and edges of the same type, and the feature type nodes are determined based on the feature types in the plant morphological response trajectory set and the stress resistance prototype trajectory pattern library. By extracting the target network and template network, subsequent node correspondence identification and edge weight comparison can be performed.

[0122] When extracting the target network and template network, node information, edge information, and edge weight information of the synergistic network are obtained from the plant stress resistance interaction graph and candidate prototypes, respectively. This information is then stored in appropriate data structures, such as adjacency matrices or adjacency lists.

[0123] Step S432: Identify the node correspondence between the target network and the template network, and sort them according to the influence priority of the feature type nodes in the resilience assessment. The influence priority is determined by the sum of the synergistic effect strength of the feature type in the pressure response sensitive range. The higher the sum, the higher the priority. Node correspondence is established in sequence.

[0124] Node correspondence identification aims to find the correspondence between nodes in the target network and the template network. Ranking nodes according to their influence priority in resilience assessment allows for a more accurate establishment of these correspondences. Influence priority is determined by the sum of the synergistic effects of each feature type within the stress response sensitivity range; a higher sum indicates greater importance and priority for that feature type during resilience assessment.

[0125] As one implementation method, step S432 can be specifically implemented as the following steps S4321~S4326: Step S4321: Calculate the sum of the synergistic effects of all incoming and outgoing edges of nodes of each feature type within the pressure response sensitive range. The intensity of the incoming edges is converted according to a preset ratio and then added to the intensity of the outgoing edges to obtain the comprehensive influence value of the node.

[0126] The overall node impact value is an indicator used to measure the importance of each feature type of node in the stress resistance process. It is calculated by summing the synergistic effects of all incoming and outgoing edges of each feature type of node within the pressure response sensitive range, taking into account the input and output relationships of the nodes. The incoming edge strength is converted according to a preset ratio and then added to the outgoing edge strength to balance the influence of incoming and outgoing edges.

[0127] When calculating the overall impact value of a node, all incoming and outgoing edges of nodes of each feature type in the target network and template network are traversed to obtain the collaborative effect strength value of each edge. For incoming edges, a preset ratio (e.g., 0.8) is applied for reduction, and then the reduced incoming edge strength is added to the outgoing edge strength to obtain the overall impact value of the node.

[0128] Step S4322: Perform interval mapping processing on the comprehensive impact value of the nodes. Based on the distribution law of the contribution of the stress resistance characteristics of plants on high-altitude and steep slopes, map the comprehensive impact value to the preset impact priority level. The level is divided into core level, important level, secondary level and related level.

[0129] Interval mapping converts the comprehensive impact value of nodes into a preset impact priority level. Based on the distribution pattern of the contribution of vegetation resistance characteristics on high-altitude and steep slopes, the range of comprehensive impact values ​​corresponding to different levels is determined. For example, after extensive experiments and data analysis, it was found that when the comprehensive impact value is in the interval [0, a), it is mapped to the correlation level; when it is in the interval [a, b), it is mapped to the secondary level; when it is in the interval [b, c), it is mapped to the importance level; and when it is in the interval [c, +∞), it is mapped to the core level, where a, b, and c are thresholds determined based on actual data.

[0130] When performing interval mapping, the comprehensive impact value of each node is compared with these intervals to determine its impact priority level. For example, using conditional statements, if the comprehensive impact value of a node is less than 'a', the node is marked as a related level; if it is greater than or equal to 'a' and less than 'b', it is marked as a secondary level, and so on. For the calculated comprehensive impact value of each node, its impact priority level is determined according to the above interval mapping rules.

[0131] Step S4323: Identify the node correspondence in the order of core level, importance level, secondary level, and association level. Core level nodes are given priority for full attribute matching, including feature type identifier, typical cooperative action mode, and degree centrality parameter in the network.

[0132] Node correspondence identification is performed in order of priority, with core nodes, due to their highest importance, being prioritized for full attribute matching. Full attribute matching requires that core nodes in the target and template networks match in three aspects: feature type identification, typical cooperative action patterns, and degree centrality parameters within the networks.

[0133] Feature type identifiers are fundamental attributes of nodes, used to distinguish different types of feature nodes, such as directional component nodes of morphological change trajectories during growth stages, and timestamp nodes of inflection point features in stress response. Typical cooperative interaction patterns refer to the ways and rules of cooperation between a node and other nodes, such as which nodes a node has strong cooperative relationships with, and the trend of changes in the strength of cooperative interactions. Degree centrality is an indicator of a node's importance in a network, including the node's in-degree (the number of edges pointing to the node) and out-degree (the number of edges emanating from the node).

[0134] When performing full attribute matching of core-level nodes, the core-level nodes in the target network and the template network are traversed. For each pair of nodes, their feature type identifiers are compared to see if they are the same, their typical cooperative action patterns are similar, and their degree centrality parameters are equal within a certain error range. For example, for feature type identifiers, a direct string comparison is performed; for typical cooperative action patterns, the similarity between them and the cooperative action strength sequences of other nodes can be compared; for degree centrality parameters, the absolute value of their difference is calculated, and if it is less than a preset error threshold, the pair is considered a match.

[0135] Step S4324: Verify the network module division of the core node matching results. Identify the tightly connected modules to which the core nodes in the target network and the template network belong using the community detection algorithm. When the module structure similarity exceeds the preset module threshold, confirm the correspondence of the core nodes.

[0136] Network module partitioning verification is to further ensure the accuracy of the correspondence between core-level nodes. The community detection algorithm is used to partition the network into different tightly connected modules, which can identify the module to which the core nodes in the target network and the template network belong.

[0137] Commonly used community detection algorithms, such as the Louvain algorithm, work by iteratively merging nodes into different communities to maximize the connection density within communities and minimize the connection density between communities. This algorithm is used to partition the target network and the template network into modules, identifying the modules to which the core nodes belong.

[0138] Module structure similarity is an indicator that measures the degree of similarity between two modules. It can be calculated by comparing the node composition and edge connections of the two modules. For example, the proportion of identical nodes and edges in the two modules can be calculated, and these proportions can be weighted and summed to obtain the module structure similarity. When the module structure similarity exceeds a preset module threshold (e.g., 0.8), the modules to which the core nodes belong are considered to have similar structures, and the correspondence between the core nodes is confirmed.

[0139] Step S4325: The important node matching is based on the correspondence of core nodes and performs local neighborhood matching. The shortest path length between the matched node and the confirmed core node is limited to no more than a preset path value. At the same time, the ratio of the cooperative effect strength between the nodes is verified to be within a preset ratio range.

[0140] Importance node matching is performed after the correspondence between core nodes has been determined. Local neighborhood matching can more accurately establish the correspondence between importance nodes. Local neighborhood refers to the range of nodes that are close to the confirmed core nodes. Limiting the shortest path length between the matching node and the confirmed core node to no more than a preset path value (e.g., 3) can narrow down the matching range and improve the accuracy of matching.

[0141] Meanwhile, verifying that the ratio of the cooperation strength between nodes is within a preset range ensures that the cooperation relationships between nodes are similar. For example, for important node A in the target network and candidate matching node B in the template network, the cooperation strength between them and the confirmed core node is calculated, and then the ratio of these two cooperation strengths is calculated. If this ratio is within a preset range (such as [0.8, 1.2]), then the cooperation relationship between the nodes is considered to be matched.

[0142] When performing local neighborhood matching for important nodes, firstly, unmatched important nodes in the target network are identified. Then, nodes in the template network whose shortest path length to the confirmed core nodes does not exceed a preset path value are selected as candidate matching nodes. For each pair of candidate nodes, the ratio of their cooperative strength with the confirmed core nodes is calculated, and the nodes are filtered according to a preset ratio range to determine the final matching nodes.

[0143] Step S4326: Secondary and related nodes adopt a greedy matching strategy. Under the premise of type consistency, the node with the smallest difference in edge weight with the confirmed node is matched first, and a complete node correspondence table is generated.

[0144] Greedy matching strategy is a strategy that selects the current optimal solution at each step. For secondary and related nodes, under the premise of type consistency, it prioritizes matching the node with the smallest difference in edge weight between the node and the confirmed node.

[0145] During matching, the secondary and related nodes in the target network are traversed. For each node, nodes with the same feature type are searched in the template network as candidate matching nodes. Then, the difference between the edge weights of each candidate matching node and the confirmed node is calculated, and the node with the smallest difference is selected as the matching node.

[0146] The difference in edge weights can be represented by calculating the absolute value of the difference in weights between corresponding edges. For example, if node A in the target network has an edge with the confirmed node C and the edge weight is w1, and candidate matching node B in the template network also has an edge with the corresponding confirmed node D and the edge weight is w2, then the difference in edge weights is |w1-w2|.

[0147] The information of each pair of matching nodes is recorded in a node correspondence table, which contains the target network nodes, template network nodes, and their correspondence.

[0148] Step S433: Based on the node correspondence, perform direction consistency verification on the directed edges of the target network and the template network, and use the ratio of the number of edges with the same direction to the total number of edges in the template network as the direction consistency rate.

[0149] Directional consistency checking compares whether the directions of directed edges in the target network and the template network are consistent. Based on the established node correspondences, for each pair of corresponding nodes in the target network and the template network, the direction of the directed edges between them is checked.

[0150] During validation, the node mapping table is traversed. For each pair of corresponding nodes, the direction of the edge between these two nodes in the target network is checked against the direction of the edge between the corresponding nodes in the template network. If they are the same, the edge is considered to have the same direction, and the number of edges with the same direction is counted.

[0151] The direction consistency rate is the ratio of the number of edges with the same direction to the total number of edges in the template network, reflecting the similarity of the directed edge directions of the two networks. For example, if the template network has N total edges and M edges with the same direction, then the direction consistency rate is M / N. For both the target network and the template network, direction consistency is checked based on the node correspondence, and the direction consistency rate is calculated.

[0152] Step S434: Calculate the degree of difference in the weight values ​​of the corresponding edges. The degree of difference is represented by the ratio of the absolute difference between the target network edge weight and the template network edge weight to the template network edge weight. The smaller the ratio, the higher the weight consistency.

[0153] The degree of difference in the weight values ​​of corresponding edges is an indicator used to measure the similarity of the weight values ​​of corresponding edges in the target network and the template network. Based on the node correspondence, corresponding edges in the target network and the template network are identified, and the degree of difference in their weight values ​​is calculated.

[0154] For each pair of corresponding edges, let the edge weight of the target network be w. t1 The edge weights of the template network are w. t2 Then the formula for calculating the degree of difference in weight values ​​is: (|w t1 -w t2 | / wt2 )×100%.

[0155] By calculating the degree of difference in weight values ​​for all corresponding edges, a sequence of differences can be obtained, which reflects the overall similarity of edge weights between the two networks. For corresponding edges in the target network and the template network, the degree of difference in weight values ​​is calculated using the formula described above.

[0156] Step S435: The directional consistency rate and weight consistency are weighted and fused. When the weight is fused, the proportion of the directional consistency rate is adjusted according to the priority of the node correspondence. The higher the priority of the node correspondence, the higher the proportion of the directional consistency rate of its connecting edge, and the network structure similarity is generated.

[0157] Weighted fusion combines directional consistency rate and weight consistency to obtain a comprehensive network structure similarity index. The proportion of directional consistency rate during weighted fusion is adjusted according to the priority of node correspondences, because higher-priority node correspondences have a greater impact on network structure similarity.

[0158] For core-level node relationships, the direction consistency rate of its connecting edges can be set to a relatively high value, such as 0.7; for important-level node relationships, the direction consistency rate can be set to 0.6; for secondary-level node relationships, the rate is 0.5; and for related-level node relationships, the rate is 0.4. Let the direction consistency rate be R. d Weight consistency can be represented by subtracting the degree of difference in the average weight values ​​from 1, denoted as R. w For each pair of corresponding nodes, the proportion α of directional consistency rate is determined according to their priority. Then, the network structure similarity S can be expressed by the formula S = α × R. d +(1-α)×R w Calculation. When calculating the network structure similarity of the entire network, a weighted average of the network structure similarity of all corresponding nodes is performed. The weights can be determined based on the importance (i.e., priority) of the nodes.

[0159] Step S436: Extract the self-adjusting loop strength of each feature type node in the network. The self-adjusting loop strength is the weight value of the directed edge pointing to itself. Calculate the relative deviation of the self-adjusting loop strength of the corresponding nodes in the target network and the template network. The relative deviation is represented by the ratio of the difference between the target loop strength and the template loop strength to the template loop strength.

[0160] The strength of the self-regulating loop reflects the intensity of a node's hysteresis effect on itself. By extracting the strength of the self-regulating loop of each feature type node in the target network and the template network and calculating their relative deviation, the similarity between the two networks can be further measured.

[0161] When extracting the self-adjusting cycle strength, each node in both the target network and the template network is traversed, and it is checked whether there is a directed edge pointing to itself. If so, the weight value of the edge is recorded as the self-adjusting cycle strength. For each pair of corresponding nodes, let w be the self-adjusting cycle strength of the node in the target network. t1_self The strength of the self-regulating cycle of the corresponding node in the template network is w. t2_self The formula for calculating the relative deviation is: ((w t1_self -w t2_self ) / w t2_self )×100%.

[0162] Step S437: When the relative deviation is less than the preset deviation threshold, increase the network structure similarity by a preset ratio; otherwise, keep the network structure similarity unchanged and generate the final network matching degree.

[0163] The preset deviation threshold is a value set in advance based on actual needs or experience. It is used to determine whether the relative deviation of the self-regulation loop strength of the corresponding nodes in the target network and the template network is within an acceptable range. When the relative deviation is less than the preset deviation threshold (e.g., 10%), it indicates that the self-regulation mechanisms of the corresponding nodes in the two networks are highly similar, and the network structure similarity is increased by a preset ratio (e.g., 5%).

[0164] When making a judgment, the relative deviation of the self-adjustment loop strength of all corresponding nodes is traversed. For nodes with a relative deviation less than a preset deviation threshold, their corresponding network structure similarity is multiplied by (1 + preset ratio); for nodes with a relative deviation greater than or equal to the preset deviation threshold, their network structure similarity remains unchanged.

[0165] Finally, the network structure similarity of all nodes is summarized and averaged to obtain the final network matching degree.

[0166] Step S440: Adjust the matching weights of the adjusted sequence and the prototype weight sequence based on the network matching degree, generate the weight sequence similarity through sequence similarity measurement, and fuse the network matching degree and the weight sequence similarity to obtain the comprehensive prototype matching degree.

[0167] Adjusting the weights based on network matching degree is done by matching the target network's weight sequence with the prototype weight sequence. A higher network matching degree indicates a greater similarity between the target and template networks, thus requiring a corresponding increase in the matching weights between the adjusted and prototype weight sequences. Sequence similarity metrics are methods used to measure the similarity between two sequences, and this method can be used to calculate the weight sequence similarity. The comprehensive prototype matching degree is a holistic index obtained by fusing network matching degree and weight sequence similarity, used to more comprehensively evaluate the matching degree between the candidate prototype and the target plant.

[0168] In one implementation, step S440 can be specifically implemented as the following steps S441~S446: Step S441: Obtain the mapping relationship between network matching degree and weight allocation coefficient. The network matching degree is converted into weight allocation coefficient through a preset piecewise function. The piecewise function sets different coefficient growth slopes according to the value range of the network matching degree. The value range of the network matching degree is divided into low matching interval, medium matching interval and high matching interval, and different intervals correspond to different coefficient growth slopes.

[0169] The mapping relationship between network matching degree and weight allocation coefficient is a rule that converts network matching degree into weight allocation coefficient, implemented through a predefined piecewise function. For example, when the network matching degree is in the low matching interval [0, 0.3), the relationship between the weight allocation coefficient y and the network matching degree x is y = 0.1x + 0.1, with a coefficient growth slope of 0.1; when it is in the medium matching interval [0.3, 0.7), the relationship is y = 0.3x + 0.04, with a coefficient growth slope of 0.3; and when it is in the high matching interval [0.7, 1], the relationship is y = 0.6x - 0.18, with a coefficient growth slope of 0.6.

[0170] When obtaining the mapping relationship, the interval to which it belongs is determined by using conditional statements based on the network matching degree value, and then the corresponding piecewise function is substituted to calculate the weight allocation coefficient.

[0171] Step S442: Divide the weight adjustment sequence and the prototype weight sequence into subsequences within the sensitive interval and subsequences within the non-sensitive interval. The subsequences within the sensitive interval are weighted using weight allocation coefficients determined by the mapping relationship, while the subsequences within the non-sensitive interval are weighted using basic coefficients, which are preset fixed values.

[0172] Dividing the weighted sequence from the original weighted sequence allows for different weighting processes based on the characteristics of different intervals. Subsequences within sensitive intervals have a greater impact on plant stress resistance, and weighting with weight allocation coefficients determined by mapping relationships can more accurately reflect their importance. Subsequences within non-sensitive intervals are weighted with basic coefficients, which are preset fixed values ​​to ensure relatively stable weights within those intervals.

[0173] During the partitioning process, based on the start and end positions of the pressure response sensitive interval, the weighted sequence and the prototype weighted sequence are divided into subsequences within the sensitive interval and subsequences within the insensitive interval, respectively. For subsequences within the sensitive interval, each element is multiplied by the weight allocation coefficient obtained through network matching degree; for subsequences within the insensitive interval, each element is multiplied by the base coefficient. For example, if the weighted sequence is [w1,w2,w3,w4,w5], and the pressure response sensitive interval is the 2nd to 4th element, then the subsequence within the sensitive interval is [w2,w3,w4], and the subsequence within the insensitive interval is [w1,w5]. Assuming the weight allocation coefficient obtained through network matching degree is 0.5 and the base coefficient is 0.2, then the weighted subsequence within the sensitive interval is [0.5w2,0.5w3,0.5w4], and the subsequence within the insensitive interval is [0.2w1,0.2w5].

[0174] Step S443: Perform multi-scale sliding window similarity measurement on the weighted target sequence and the prototype sequence. The window size is determined according to the total number of growth stages. Each window covers multiple consecutive growth stages. The window sliding step size is one growth stage. Calculate the cosine similarity of sequence elements within each window to generate a multi-scale similarity set containing local and global similarities. The local similarity is the average of the cosine similarities of each window, and the global similarity is the cosine similarity of the entire sequence.

[0175] Multi-scale sliding window similarity measurement is a method for comparing the similarity of two sequences. By sliding windows of different sizes across the sequences, the cosine similarity of sequence elements within the window is calculated, providing similarity information at different scales. The window size is determined by the total number of growth stages, with each window covering multiple consecutive growth stages. The window sliding step size is one growth stage, ensuring a comprehensive analysis of sequence similarity.

[0176] As one implementation method, step S443 can be specifically implemented as the following steps S4431~S4436: Step S4431: The product of the square root of the total number of growth stages and the length of the pressure response sensitive interval is used as the basis for calculating the basic window size. The basic window size increases non-linearly with the increase of the total number of growth stages and increases linearly with the increase of the sensitive interval length.

[0177] The calculation of the basic window size comprehensively considers the total number of growth stages and the length of the pressure response sensitive interval. The product of the square root of the total number of growth stages and the length of the pressure response sensitive interval ensures that the basic window size has a reasonable size under different growth cycles and sensitive interval lengths. Due to the square root calculation of the total number of growth stages, the basic window size increases non-linearly with the increase of the total number of growth stages; however, it has a linear multiplication relationship with the length of the pressure response sensitive interval, so it increases linearly with the increase of the sensitive interval length.

[0178] Step S4432: Design a window adjustment mechanism, extract the weight change rate of the weight adjustment sequence in the sequence, when the absolute value of the change rate exceeds the preset fluctuation threshold, the window size is reduced by the preset adjustment ratio, and vice versa. The adjusted window size is limited to a preset multiple of the basic window size.

[0179] The window adjustment mechanism dynamically adjusts the window size based on the weight changes in the weight adjustment sequence. The weight change rate reflects the degree of change in the weight adjustment sequence at different stages. When the absolute value of the change rate exceeds a preset fluctuation threshold (e.g., 0.1), it indicates a large weight change, and the window size is reduced by a preset adjustment ratio (e.g., 0.8) to more precisely capture the changes in the sequence; conversely, the window size is increased by an adjustment ratio (e.g., 1.2) to improve computational efficiency.

[0180] The adjusted window size is limited to a preset multiple of the base window size, for example, between 0.5 and 2 times the base window size. When calculating the weight change rate, for the weight adjustment sequence [w1, w2, ..., wn], the weight change rate is [(w2-w1) / w1, (w3-w2) / w2, ..., (wn-wn-1) / wn-1]. For each window, its corresponding weight change rate is calculated, and the window size is adjusted based on the comparison between the change rate and a preset fluctuation threshold.

[0181] Step S4433: During the window sliding process, extract the morphological features of the weight sequence. Extract the weight distribution entropy, the slope of the weight change trend, and the number of local extreme points for the sequence segment in each window as window feature descriptors.

[0182] Weight sequence morphological feature extraction aims to more comprehensively describe the features of the sequence within the window. The weight distribution entropy reflects the uniformity of the weight distribution within the window, the slope of the weight change trend reflects the trend of weight change, and the number of local extreme points reflects the fluctuation of the weight.

[0183] The entropy of the weight distribution can be obtained by calculating the entropy of the probability distribution of the weights within the window. Assuming the weight sequence within the window is [w1, w2, ..., wk], first calculate the probability of each weight value appearing pi = wi / ∑wj (j from 1 to k), and then calculate the weight distribution entropy value according to the entropy calculation formula H = -∑pi × log(pi).

[0184] The slope of the weight change trend can be obtained by performing linear regression analysis on the weight sequence within the window. The slope of the linear regression equation is the slope of the weight change trend.

[0185] The number of local extrema refers to the number of maxima and minima in the weight sequence within a window. Extrema can be determined by iterating through the sequence within the window and comparing the values ​​of each point with its neighbors.

[0186] Step S4434: Calculate the feature descriptor similarity between the target window and the prototype window, and generate a comprehensive window similarity by combining the cosine similarity. The feature descriptor similarity is represented by a weighted sum of the differences in distribution entropy, trend slope, and number of extreme points.

[0187] The window-based comprehensive similarity takes into account both feature descriptor similarity and cosine similarity. Feature descriptor similarity is represented by a weighted sum of differences in distribution entropy, trend slope, and the number of extreme points.

[0188] Let the entropy of the weight distribution of the target window be H. t1 The entropy value of the weight distribution of the prototype window is H. t2 The difference in distribution entropy values ​​is |H t1 -H t2 |;The slope of the weight change trend of the target window is k t1 The slope of the weight change trend of the prototype window is k. t2 The difference in trend slope is |k t1 -k t2 |;The number of local extrema in the target window is n t1 The number of local extrema of the prototype window is n. t2 The difference in the number of extreme points is |n t1 -n t2 |

[0189] Assuming the weights of the differences in distribution entropy, trend slope, and number of extreme points are α, β, and γ (α+β+γ=1), respectively, then the feature descriptor similarity S des =α×|H t1 -H t2 |+β×|k t1 -k t2 |+γ×|n t1 -n t2 |

[0190] Window overall similarity S win =(1-δ)×S c +δ×(1-S des ), where S c It is the cosine similarity between the sequence elements within the target window and the prototype window. δ is an adjustment coefficient used to balance the influence of cosine similarity and feature descriptor similarity.

[0191] Step S4435: The global similarity calculation adopts a segmented aggregation approximation algorithm, which converts the entire weight sequence into a combination of multiple linear segments. The global structural similarity is generated by comparing the segment slope, length and starting position of the target sequence and the prototype sequence.

[0192] The segmented aggregation approximation algorithm transforms the entire weight sequence into a combination of multiple linear segments, each of which can be described by its slope, length, and starting position. By comparing the slope, length, and starting position of these linear segments between the target sequence and the prototype sequence, a global structural similarity can be generated.

[0193] When performing segmented aggregation approximation, the entire weight sequence is first divided into multiple linear segments. A sliding window method can be used to perform linear fitting within each window, obtaining the slope, length, and starting position of the linear segment within that window. Then, for each linear segment of the target sequence and the prototype sequence, their slope, length, and starting position are compared.

[0194] For example, for slopes, the absolute value of their differences can be calculated. If the difference is less than a preset slope error threshold, the slopes are considered to match. For length and starting position, corresponding error thresholds can also be set for matching. The global structural similarity is calculated based on the ratio of the number of matched segments to the total number of segments.

[0195] Step S4436: The multi-scale similarity set includes the window comprehensive similarity matrix under different window sizes, the global structural similarity, and the feature descriptor matching results of each window. The row dimension of the matrix is ​​the window position, and the column dimension is the window size.

[0196] The multi-scale similarity set integrates the window comprehensive similarity matrix under different window sizes, global structural similarity, and feature descriptor matching results for each window. The window comprehensive similarity matrix records the window comprehensive similarity values ​​under different window positions and window sizes; the row dimension of the matrix represents the window position, and the column dimension represents the window size.

[0197] Global structural similarity is the similarity of the entire sequence calculated using a segmented aggregation approximation algorithm. The feature descriptor matching results for each window record information such as the feature descriptor similarity for each window.

[0198] When generating a multi-scale similarity set, the window comprehensive similarity matrix, global structural similarity, and feature descriptor matching results of each window calculated under different window sizes are integrated into a data structure, such as a multi-dimensional array or a dictionary.

[0199] Step S444: Perform a weighted summation on the multi-scale similarity set, with the window similarity weight corresponding to the sensitive interval being a preset multiple of the window similarity weight of the non-sensitive interval, to generate the final weighted sequence similarity.

[0200] Weighted summation combines the similarities of each item in the multi-scale similarity set to obtain a comprehensive weighted sequence similarity. The window similarity corresponding to the sensitive interval has a significant impact on the plant's stress resistance, and its weight is a preset multiple (e.g., 2 times) of the weight of the window similarity of the non-sensitive interval.

[0201] When performing weighted summation, the window positions corresponding to sensitive and non-sensitive intervals are first determined. For windows corresponding to sensitive intervals, their similarity weights are multiplied by a preset factor; for windows corresponding to non-sensitive intervals, the default weights are used. Then, the window comprehensive similarity matrix, global structural similarity, and feature descriptor matching results of each window in the multi-scale similarity set are weighted and summed.

[0202] For example, for the window-based comprehensive similarity matrix, each element is multiplied by its corresponding weight, and then all elements are summed; for the global structural similarity and feature descriptor matching results, they are also multiplied by their respective weights and then summed. The final sum is the final weighted sequence similarity.

[0203] When fusing network matching degree and weight sequence similarity, a fusion rule based on network matching degree is adopted. When the network matching degree is in the high matching degree range, the proportion of network matching degree in the fusion is increased by a preset ratio, and the proportion of weight sequence similarity is decreased by a corresponding ratio. When the network matching degree is in the low matching degree range, the basic proportion of network matching degree and weight sequence similarity is maintained. When the network matching degree is in the medium matching degree range, the proportion is adjusted in a linear transition manner.

[0204] The fusion rule based on network matching degree adjusts the proportion of network matching degree and weight sequence similarity in the fusion process according to the level of network matching degree, so as to more reasonably combine these two indicators.

[0205] Step S445: Calculate the overall prototype matching degree through the fusion rules, compare the overall prototype matching degrees of all candidate prototypes, and select the prototype corresponding to the highest value to enter the reliability verification stage of the matching result.

[0206] The comprehensive prototype matching degree is a comprehensive index obtained by fusing network matching degree and weight sequence similarity. By comparing the comprehensive prototype matching degree of all candidate prototypes, the prototype corresponding to the highest value is selected to enter the reliability verification stage of the matching result, which can ensure that the final selected prototype has the highest degree of matching with the target plant.

[0207] During the comparison, the overall prototype matching score of all candidate prototypes is stored in a list. The list is then sorted using a sorting algorithm (such as bubble sort or quicksort), and the prototype corresponding to the maximum value is selected. For example, for a candidate prototype set [P1, P2, ..., Pn], their overall prototype matching scores are [S1, S2, ..., Sn]. After sorting, the prototype with the highest overall prototype matching score is selected, let's say Pk. For all candidate prototypes, the prototype corresponding to the highest overall prototype matching score is selected and proceeded to the reliability verification stage of the matching results, following the method described above.

[0208] Step S450: Verify the reliability of the matching results for the candidate prototype with the highest overall prototype matching degree. After verifying the consistency of feature contribution distribution, output the corresponding resilience level as the classification result.

[0209] The reliability verification of the matching results is to ensure that the matching results between the finally selected candidate prototype and the target plant are reliable. The consistency verification of the feature contribution distribution can check whether the candidate prototype and the target plant are consistent in the contribution distribution of each feature.

[0210] The consistency verification of feature contribution distribution can be performed by comparing the distribution of contribution weights of the candidate prototype and the target plant on each feature. For example, the correlation coefficient of the contribution weights of the two can be calculated. If the correlation coefficient exceeds a preset threshold (e.g., 0.8), the feature contribution distribution is considered to be consistent.

[0211] During validation, the contribution weights of each feature of the candidate prototype and the target plant are stored in two separate vectors, and their correlation coefficients are calculated using the correlation coefficient calculation formula. If the validation passes, the stress resistance level corresponding to the candidate prototype is output as the classification result, such as high stress resistance, medium stress resistance, low stress resistance, etc.

[0212] Please see Figure 2 , Figure 2This is a schematic diagram of a computer system provided in an embodiment of the present invention. The computer system includes at least a processor 101, a communication interface 102, and a memory 103. The processor 101, communication interface 102, and memory 103 can be connected via a bus or other means. The processor 101 (or Central Processing Unit, CPU) is the computing and control core of the computer system, capable of parsing various instructions and processing various data within the computer system. The communication interface 102 may optionally include a standard wired interface or a wireless interface (such as Wi-Fi, mobile communication interface, etc.), and can be used to send and receive data under the control of the processor 101; the communication interface 102 can also be used for data transmission and interaction within the computer system. The memory 103 is a storage device in the computer system used to store programs and data. It is understood that the memory 103 here can include the computer system's built-in memory, or it can include extended memory supported by the computer system. The memory 103 provides storage space, which stores the computer system's operating system; this invention does not limit this storage space.

[0213] In one embodiment, the processor 101 executes the method for classifying stress-resistant plants on cold and steep slopes based on plant image processing provided in the above embodiments of the present invention by running a computer program in the memory 103.

Claims

1. A classification method for stress-resistant plants on cold and steep slopes based on plant image processing, characterized in that, The method includes: Collect time-series growth morphology images of target plant populations in high-altitude and steep slope areas within a preset growth cycle. The time-series growth morphology images are sequences of overall plant morphological changes that include multiple consecutive growth stages. The temporal growth morphology images are subjected to morphological response trajectory analysis processing. By extracting the morphological change vectors from the inter-frame difference of the temporal images, a set of plant morphological response trajectories containing morphological change trajectories of growth stages, inflection point features of pressure response, and temporal correlation features are obtained. Based on the set of plant morphological response trajectories, stress resistance interaction modeling is performed, and a plant stress resistance interaction map is generated by calculating the intensity of multi-feature synergistic effects and adjusting the stress resistance contribution. The plant morphological response trajectory set and the plant stress resistance interaction map are progressively matched with a preset stress resistance prototype trajectory pattern library to determine the classification results of stress-resistant plants on cold and steep slopes.

2. The method for classifying stress-resistant plants on steep, cold slopes based on plant image processing according to claim 1, characterized in that, The process of parsing the morphological response trajectory of the time-series growth morphology images, through inter-frame difference analysis and morphological change vector extraction of the time-series images, yields a set of plant morphological response trajectories containing morphological change trajectories of growth stages, stress response inflection point features, and time-series correlation features, including: The temporal growth morphology image is subjected to temporal alignment processing. Based on the image acquisition timestamp, image frames of multiple growth stages are calibrated into a temporal image sequence with equal time intervals to generate a temporally aligned image sequence. The temporally aligned image sequence is subjected to inter-frame difference processing to calculate the pixel grayscale difference between adjacent image frames. The inter-frame morphological change region is extracted by threshold segmentation to generate a binarized image sequence of the morphological change region. Morphological dilation and erosion are performed on each binarized image in the binarized image sequence of the morphological change region to eliminate noise interference and connect the broken change regions, thereby generating a denoised image of the morphological change region. Extract the centroid coordinate sequence of the morphological change region from the denoised morphological change region image, calculate the morphological change vector based on the change of centroid coordinates of consecutive timestamps, and arrange the morphological change vectors in chronological order to form the morphological change trajectory of the growth stage. The inflection point detection process is performed on the morphological change trajectory of the growth stage. The extreme points in the trajectory are identified by calculating the rate of change of the second derivative of the trajectory curve. The timestamp and morphological change vector corresponding to the extreme points are used as the inflection point features of the pressure response. Calculate the cosine similarity of morphological change trajectories of adjacent growth stages, generate temporal correlation features between trajectories, and unify the dimensions of the morphological change trajectories of the growth stages, the inflection point features of the pressure response, and the temporal correlation features to generate a set of plant morphological response trajectories.

3. The method for classifying stress-resistant plants on cold and steep slopes based on plant image processing according to claim 2, characterized in that, The inflection point detection process for the morphological change trajectory during the growth stage involves identifying extreme points in the trajectory by calculating the rate of change of the second derivative of the trajectory curve, and using the timestamp and morphological change vector corresponding to the extreme points as features of the pressure response inflection point. This includes: The morphological change trajectory of the growth stage is represented as a trajectory function with timestamp as the independent variable, and the dependent variables of the trajectory function are the direction component and the magnitude component of the morphological change vector. The second derivatives of the direction component and the magnitude component of the trajectory function are respectively performed to obtain the second derivative sequences of the direction and the second derivative sequences of the magnitude. Calculate the absolute value of the difference between adjacent elements in the second derivative sequence of the direction. When the absolute value of the difference exceeds a preset direction change threshold, mark the corresponding timestamp as a candidate inflection point of the direction. Calculate the absolute value of the difference between adjacent elements in the second derivative sequence of the modulus length. When the absolute value of the difference exceeds the preset modulus length change threshold, mark the corresponding timestamp as a candidate inflection point of the modulus length. Perform a timestamp intersection operation on the candidate inflection points of direction and candidate inflection points of magnitude, and determine the timestamp that simultaneously satisfies the thresholds of both direction and magnitude change as the pressure response inflection point; Extract the morphological change vector corresponding to the pressure response inflection point, including the angle value of the directional component and the length ratio of the modulus component, and combine them to form the pressure response inflection point feature.

4. The method for classifying stress-resistant plants on steep, cold slopes based on plant image processing according to claim 1, characterized in that, The process of modeling stress resistance interactions based on the set of plant morphological response trajectories, and generating a plant stress resistance interaction map through multi-feature synergistic effect strength calculation and stress resistance contribution adjustment, includes: Extract the morphological change trajectory of the growth stage, the inflection point feature of the pressure response, and the temporal correlation feature from the set of plant morphological response trajectories, and construct a three-dimensional feature interaction matrix. The row dimension of the matrix is ​​the growth stage, the column dimension is the feature type, and the depth dimension is the feature value. The three-dimensional feature interaction matrix is ​​subjected to pressure-sensitive interval identification processing. By scanning the distribution of pressure response inflection points in the matrix through a sliding window, the continuous growth stage where the feature interaction intensity exceeds a preset threshold is determined and marked as the pressure response sensitive interval. Within the pressure response sensitive range, the strength of multi-feature synergy is calculated. A synergy network is constructed by the mutual information entropy and partial correlation coefficient between features. The network nodes are feature types and the edge weights are the synergy strengths. The resilience contribution weights of each feature are adjusted based on the edge weights of the synergistic network. The higher the synergistic strength of a feature combination, the greater the increment of the corresponding feature's contribution weight, thus generating a weight adjustment sequence. The three-dimensional feature interaction matrix, pressure response sensitive region, synergistic network and weight adjustment sequence are integrated into a graph, and the feature interaction direction is represented by a directed graph, with the node size representing the contribution weight, to generate a plant stress resistance interaction graph.

5. The method for classifying stress-resistant plants on cold and steep slopes based on plant image processing according to claim 4, characterized in that, The step of performing pressure-sensitive interval identification processing on the three-dimensional feature interaction matrix involves scanning the distribution of pressure response inflection points in the matrix through a sliding window to determine continuous growth stages where the feature interaction intensity exceeds a preset threshold, and marking these as pressure-sensitive intervals. Set the time sliding window size to the preset number of growth stages, and slide it along the row dimension of the three-dimensional feature interaction matrix with a window step size as one growth stage. Calculate the density value of the pressure response inflection point within each window, and represent the density value as the ratio of the number of inflection points to the window size; Extract the interaction covariance matrix of all feature types within the window, and calculate the trace of the covariance matrix as an indicator of feature interaction strength. When the inflection point density value of a window exceeds the density threshold and the feature interaction intensity index exceeds the intensity threshold, the window is marked as a candidate sensitive region. Adjacent candidate sensitive intervals are merged to form pressure response sensitive intervals in a continuous growth stage. The start of the interval is the starting stage of the first candidate window, and the end of the interval is the ending stage of the last candidate window.

6. The method for classifying stress-resistant plants on steep, cold slopes based on plant image processing according to claim 4, characterized in that, The calculation of the multi-feature synergistic effect strength within the pressure response sensitive range involves constructing a synergistic effect network using the mutual information entropy and partial correlation coefficient between features. Network nodes represent feature types, and edge weights represent the synergistic effect strength. This includes: Within the pressure response sensitive range, five feature types are extracted from the three-dimensional feature interaction matrix: the directional component of the morphological change trajectory during the growth stage, the modulus component, the timestamp of the pressure response inflection point feature, the vector value, and the similarity value of the temporal correlation feature. Calculate the mutual information entropy between any two feature types. The mutual information calculation measures the degree of non-linear correlation between features. The larger the mutual information entropy value, the stronger the non-linear correlation. Calculate the partial correlation coefficient between any two feature types. While keeping other feature types constant, measure the degree of linear association between the two features. The larger the absolute value of the partial correlation coefficient, the stronger the linear association. The mutual information entropy value and the absolute value of the partial correlation coefficient are weighted and fused to generate the synergistic effect strength value; Using five feature types as network nodes and the cooperative effect strength value as the weight of the directed edge between nodes, a cooperative effect network containing self-regulating loops is constructed. The self-regulating loops represent the lag effect strength of a feature on itself.

7. The method for classifying stress-resistant plants on cold and steep slopes based on plant image processing according to claim 1, characterized in that, The step of progressively matching the set of plant morphological response trajectories and the plant stress resistance interaction map with a preset stress resistance prototype trajectory pattern library to determine the classification results of stress-resistant plants on high-altitude and steep slopes includes: Obtain a preset resilience prototype trajectory pattern library, which contains typical trajectory patterns of different resilience levels. Each typical trajectory pattern includes a prototype morphological trajectory, a prototype sensitive region, a prototype cooperative network, and a prototype weight sequence. The growth stage morphological change trajectory is extracted from the plant morphological response trajectory set, and multiple rounds of prototype screening are performed with the prototype morphological trajectory. The candidate prototype set is determined by trajectory trend matching and sensitive interval verification. For each candidate prototype in the candidate prototype set, a collaborative network structure alignment process is performed, and the network matching degree is generated by identifying the node correspondence and comparing the edge weights. The matching weights of the adjusted weight sequence and the prototype weight sequence are adjusted based on the network matching degree. The weight sequence similarity is generated by sequence similarity measurement. The comprehensive prototype matching degree is obtained by fusing the network matching degree and the weight sequence similarity. The reliability of the matching results is verified for the candidate prototype with the highest overall prototype matching degree. After verifying the consistency of feature contribution distribution, the corresponding resilience level is output as the classification result.

8. The method for classifying stress-resistant plants on steep, cold slopes based on plant image processing according to claim 7, characterized in that, The process of extracting morphological change trajectories at different growth stages from the plant morphological response trajectory set, performing multiple rounds of prototype screening with the prototype morphological trajectories, and determining the candidate prototype set through trajectory trend matching and sensitive interval verification includes: The trajectory of morphological changes during the growth stage is initially matched with the trajectory of the prototype morphology, and the consistency of the overall evolution direction of the trajectory is calculated by time warping algorithm. Based on the trajectory trend matching results, select several candidate prototypes with the highest similarity ranking as the initial candidate set. The size of the initial candidate set is determined according to the total number of prototypes in the resilience prototype trajectory pattern library, so that the initial candidate set covers prototypes of all resilience levels, and each resilience level contains at least one prototype. The stress response sensitive intervals are extracted from the plant stress resistance interaction map and their compatibility is verified with the prototype sensitive intervals of each prototype in the initial candidate set. This includes calculating the interval start stage difference, interval end stage difference, and interval length ratio. The interval start stage difference is the difference between the start stage of the target sensitive interval and the start stage of the prototype sensitive interval. The interval end stage difference is the difference between the end stage of the target sensitive interval and the end stage of the prototype sensitive interval. The interval length ratio is the ratio of the length of the target sensitive interval to the length of the prototype sensitive interval. When the interval start stage difference is less than the preset stage difference threshold, the interval end stage difference is less than the preset stage difference threshold, and the interval length ratio is within the preset ratio range, the candidate prototype is determined to pass the interval adaptability check; otherwise, it is removed from the initial candidate set. The prototypes that pass the interval fit test are compared with the temporal correlation features. The overall similarity of the trajectory correlation feature sequence between the target plant and the candidate prototype is calculated. The similarity is represented by the mean absolute error of the correlation feature values ​​of the corresponding stages in the sequence. The mean absolute error is calculated by the arithmetic mean of the absolute values ​​of the differences of the correlation feature values ​​of each stage in the sequence. Finally, the set of candidate prototypes that meet the multiple screening conditions is determined.

9. The method for classifying stress-resistant plants on cold and steep slopes based on plant image processing according to claim 8, characterized in that, The step of performing collaborative network structure alignment processing on each candidate prototype in the candidate prototype set, and generating network matching degree by identifying node correspondence and comparing edge weights, includes: The synergistic network is extracted from the plant stress resistance interaction map as the target network, and the prototype synergistic network is extracted from the candidate prototype as the template network. Both networks contain feature type nodes, directed edges and edge weights. Feature type nodes include directional component nodes and modulus component nodes of the morphological change trajectory of the growth stage, timestamp nodes and vector value nodes of the stress response inflection point feature, and similarity value nodes of the temporal correlation feature. The node correspondence between the target network and the template network is identified, and the nodes are ranked according to the influence priority of the feature type nodes in the stress resistance assessment. The influence priority is determined by the sum of the synergistic effect strength of the feature type in the stress response sensitive range. The higher the sum, the higher the priority. The node correspondence is established in sequence. Based on the node correspondence, the directed edges of the target network and the template network are checked for directional consistency. The ratio of the number of edges with the same direction to the total number of edges in the template network is used as the directional consistency rate. Calculate the degree of difference in the weight values ​​of the corresponding edges. The degree of difference is represented by the ratio of the absolute difference between the weight of the target network edge and the weight of the template network edge to the weight of the template network edge. The smaller the ratio, the higher the weight consistency. The directional consistency rate and weight consistency are weighted and fused. The proportion of directional consistency rate during weight fusion is adjusted according to the priority of the node correspondence. The higher the priority of the node correspondence, the higher the proportion of directional consistency rate of its connecting edge, thus generating network structure similarity. Extract the self-adjusting loop strength of nodes of each feature type in the network. The self-adjusting loop strength is the weight value of the directed edge pointing to itself. Calculate the relative deviation of the self-adjusting loop strength of the corresponding nodes in the target network and the template network. The relative deviation is represented by the ratio of the difference between the target loop strength and the template loop strength to the template loop strength. When the relative deviation is less than the preset deviation threshold, the network structure similarity is increased by a preset ratio; otherwise, the network structure similarity remains unchanged, and the final network matching degree is generated.

10. A computer system, characterized in that, include: A memory, wherein a computer program is stored; A processor is configured to load the computer program to implement the method for classifying stress-resistant plants on cold and steep slopes based on plant image processing as described in any one of claims 1-9.

Images