Classification of degradation grades and spatial early warning method and system for populations of allium multiflorum
By combining robust covariance estimation and row sparse principal component analysis with spatial smoothing regularization, the problems of outlier sensitivity and spatial discontinuity in the degradation assessment of *Allium chinense* populations in alpine grasslands are solved, enabling efficient and interpretable identification and early warning of degraded patches.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- GUIZHOU ACAD OF FORESTRY SCI
- Filing Date
- 2026-06-01
- Publication Date
- 2026-07-03
AI Technical Summary
Traditional methods for assessing the degradation of alpine meadow chive populations are sensitive to outliers and cannot effectively utilize spatial neighborhood information, resulting in spatially discontinuous assessment results and difficulty in explaining the dominant degradation factors.
Robust covariance estimation, row sparse principal component analysis, and spatial smoothing regularization are integrated into a unified optimization framework. A local spectral matrix is constructed by an iterative reweighted minimum covariance determinant method. Combined with distributed iterative optimization and a configurable rule engine, degradation types are identified and spatially continuous patches are output.
It effectively suppresses outlier interference, improves the reliability and stability of assessment results, clearly determines the type of degradation stress, and enhances the practical value and timeliness of spatial early warning layers.
Smart Images
Figure CN122332827A_ABST
Abstract
Description
Technical Field
[0001] The embodiments of this invention relate to the field of ecological environment monitoring technology, and in particular to a method and system for classifying and spatially warning about the degradation levels of *Leekia spicata* populations in alpine grasslands. Background Technology
[0002] Alpine grassland ecosystems are fragile, and the degradation of *Allium chinense*, a unique dominant species and core landscape resource in the high-altitude regions of Guizhou, has attracted widespread attention. Traditional degradation assessments rely heavily on field quadrat surveys and empirical grading, making it difficult to achieve large-scale, high-precision, continuous spatial monitoring. In recent years, methods based on remote sensing and spatial data analysis have gradually emerged, primarily relying on principal component analysis (PCA) and its variants for multi-index dimensionality reduction and degradation feature extraction. However, existing technologies still have the following significant shortcomings:
[0003] Sensitivity to outliers: Field survey data often produce outliers due to instrument errors, occasional extreme events, etc. Classical PCA and ordinary sparse PCA are based on the sample covariance matrix and are extremely sensitive to outliers.
[0004] Lack of interpretability: The principal components of traditional PCA are linear combinations of all the original variables, making it difficult to clearly identify the dominant factors driving the degradation.
[0005] Spatial discontinuity and fragmentation: The pixel-by-pixel independent analysis method ignores the similarity of spatial neighborhoods, resulting in spatially discrete and fragmented evaluation results.
[0006] Unable to utilize spatial neighborhood information: Existing technologies often treat each quadrat as an isolated sample, failing to effectively utilize surrounding information.
[0007] Therefore, there is an urgent need to develop a system that can simultaneously resist outlier interference, automatically extract interpretable key degradation variables, and output a classification and early warning system for spatially continuous degradation patches. Summary of the Invention
[0008] The purpose of this invention is to overcome the shortcomings of existing technologies and provide a method and system for classifying and spatially warning about the degradation levels of *Allium chinense* populations in alpine grasslands. By integrating robust covariance estimation, row sparse principal component analysis, and spatial smoothing regularization into a unified optimization framework, this invention enables continuous spatial assessment and classification of the degradation status of large-scale *Allium chinense* populations. It automatically outputs the boundaries, degradation levels, and dominant degradation types of degradation hotspots, providing a scientific basis and technical support for the precise protection and differentiated ecological restoration of characteristic plant populations in alpine grasslands.
[0009] In a first aspect, embodiments of the present invention provide a method for classifying and spatially warning of degradation levels of *Allium chinense* populations in alpine grasslands, including:
[0010] The target region is divided into multiple spatial grid points, and the multi-star degradation-related feature variables of each spatial grid point are obtained to construct the feature vector of each spatial grid point.
[0011] Define the spatial neighborhood information of each spatial grid point, and obtain the local spectral matrix of each spatial grid point through the robust covariance estimation method based on the feature vector of each spatial grid point and its spatial neighborhood information.
[0012] A unified optimization problem is constructed and solved based on the local spectral matrix of each spatial grid point. The optimal load matrix of each spatial grid point is output through distributed iterative optimization. The unified optimization problem aims to maximize the projection variance, filter key variables by row sparsity constraints, and ensure the principal component continuity of adjacent grid points by spatial smoothness constraints.
[0013] The degradation intensity index is calculated based on the optimal load matrix and local spectrum matrix of each spatial grid point, and the dominant degradation variable is identified based on the row comprehensive load of the optimal load matrix. The degradation type of each spatial grid point is determined by a configurable rule engine.
[0014] Spatial segmentation and connectivity analysis are performed on the degradation intensity index to extract continuous degradation patches and classify degradation levels, outputting a spatial early warning layer containing patch boundaries, degradation levels, and degradation types.
[0015] In a preferred embodiment, based on the feature vectors of each spatial grid point and their spatial neighborhood information, the local spectral matrix of each spatial grid point is obtained through a robust covariance estimation method, including:
[0016] The iterative reweighted minimum covariance determinant method is used to select the optimal subset from the spatial neighborhood of each spatial grid point, where the sample size reaches a preset proportion threshold.
[0017] The covariance matrix of the optimal subset is used as the local spectral matrix of the spatial grid point.
[0018] As a preferred embodiment, when defining the spatial neighborhood information of each spatial grid point, for spatial grid points located at the boundary of the target region, the spatial neighborhood of their spatial grid points is expanded to the full window size using a reflection filling method.
[0019] When the actual sample size of the expanded spatial neighborhood is lower than the preset sample size threshold, a data credibility warning message is marked in the output result.
[0020] In a preferred embodiment, a unified optimization problem is constructed and solved based on the local spectral matrices of each spatial grid point, and the optimal load matrix of each spatial grid point is output through distributed iterative optimization, including:
[0021] Initialize the load matrix of each spatial grid point, decompose the unified optimization problem into parallel subproblems of each spatial grid point using the inexact alternating direction multiplier method, and introduce auxiliary variables and dual variables;
[0022] The parallel subproblems of each spatial grid point are solved iteratively. In each iteration, the load matrix, auxiliary variables and dual variables of each spatial grid point are updated in parallel until the preset convergence condition is met. The converged load matrix is then output as the optimal load matrix.
[0023] The row sparsity constraint is achieved by applying a mixture norm penalty to the load matrix, and the spatial smoothness constraint is achieved by measuring the distance between the principal components of the subspace projection matrix of adjacent grid points.
[0024] As a preferred implementation, the auxiliary variables of each spatial grid point are updated in parallel, including:
[0025] The combination of the load matrix and dual variables in the current iteration step is taken as the near-end target point, and its projection matrix is calculated.
[0026] The projection matrix is smoothed using a graph Laplacian smoothing method to obtain a smoothed projection matrix.
[0027] The smoothed projection matrix is subjected to truncated singular value decomposition to obtain eigenvalues, which are then sorted.
[0028] The updated auxiliary variables are constructed by taking the eigenvectors corresponding to the n largest eigenvalues in the eigenvalue sequence, where n is the preset principal component dimension.
[0029] In a preferred embodiment, the degradation intensity index is calculated based on the optimal load matrix and local spectral matrix of each spatial grid point, including:
[0030] Project the optimal load matrix of each spatial grid point onto the local spectral matrix of that spatial grid point, and calculate the projection variance trace of each spatial grid point.
[0031] The projection variance trace of each spatial grid point is used as the degradation intensity index of the corresponding spatial grid point.
[0032] As a preferred implementation, the dominant degradation variable is identified based on the row composite load of the optimal load matrix, and the degradation type of each spatial grid point is determined by a configurable rule engine, including:
[0033] Calculate the row norm of each row of the optimal load matrix for each spatial grid point, use the row norm as the composite load value of that row, and identify one or more variables with the largest composite load value as the dominant degenerate variables of that spatial grid point.
[0034] Based on the identified dominant degradation variable, the corresponding degradation type is matched using the configurable rule engine;
[0035] The rule engine has several pre-defined degradation types and their corresponding dominant degradation variable condition combinations.
[0036] In a preferred embodiment, the degradation intensity index is spatially segmented and analyzed for connectivity to extract continuous degradation patches and classify them into degradation levels. A spatial warning layer containing patch boundaries, degradation levels, and degradation types is output, including:
[0037] A global distribution analysis of the degradation intensity index of each spatial grid point is performed, and a segmentation threshold is adaptively selected for binarization segmentation based on the distribution characteristics.
[0038] Connectivity analysis was performed on the binarized segmentation results to extract continuous spatially connected regions as degenerate patches.
[0039] Calculate the average degradation intensity and patch area for each degraded patch, and classify the degradation level of each spatial grid point based on the average degradation intensity and patch area;
[0040] Integrate the degradation types of the grid points corresponding to each degradation patch, and output a spatial warning layer that includes patch boundaries, degradation levels, and degradation types.
[0041] In a preferred embodiment, the step of adaptively selecting a segmentation threshold based on distribution characteristics for binarization segmentation includes:
[0042] A bimodality test was performed on the global distribution of the degradation intensity index;
[0043] If the global distribution exhibits a significant bimodal distribution, the segmentation threshold is determined using the method of maximizing inter-class variance; otherwise, a statistical threshold based on the interquartile range is used for segmentation.
[0044] Secondly, embodiments of the present invention also provide a system for classifying and spatially warning of degradation levels of *Allium chinense* populations in alpine grasslands, comprising:
[0045] The data acquisition and gridding module is used to divide the target area into multiple spatial grids and obtain the multi-star degradation-related feature variables of each spatial grid, and construct the feature vector of each spatial grid.
[0046] The local spectral matrix estimation module is used to define the spatial neighborhood information of each spatial grid point. Based on the feature vectors of each spatial grid point and its spatial neighborhood information, the local spectral matrix of each spatial grid point is obtained through a robust covariance estimation method.
[0047] The unified optimization problem solving module is used to construct and solve the unified optimization problem based on the local spectral matrix of each spatial grid point. It outputs the optimal load matrix of each spatial grid point through distributed iterative optimization. The unified optimization problem aims to maximize the projection variance, filter key variables by row sparsity constraints, and ensure the principal component continuity of adjacent grid points by spatial smoothness constraints.
[0048] The degradation type discrimination module is used to calculate the degradation intensity index based on the optimal load matrix and local spectrum matrix of each spatial grid point, identify the dominant degradation variable based on the row comprehensive load of the optimal load matrix, and determine the degradation type of each spatial grid point through a configurable rule engine.
[0049] The spatial early warning output module is used to perform spatial segmentation and connectivity analysis on the degradation intensity index, extract continuous degradation patches and classify degradation levels, and output a spatial early warning layer containing patch boundaries, degradation levels and degradation types.
[0050] Thirdly, embodiments of the present invention also provide an electronic device, the electronic device comprising:
[0051] One or more processors;
[0052] Storage device for storing one or more programs;
[0053] When the one or more programs are executed by the one or more processors, the one or more processors implement the method for classifying and spatially warning about the degradation level of alpine meadow ivy populations as described in any embodiment of the present invention.
[0054] Fourthly, embodiments of the present invention also provide a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the method for classifying and spatially warning the degradation levels of alpine meadow ivy populations as described in any embodiment of the present invention.
[0055] Compared to existing technologies, the following beneficial effects have been achieved:
[0056] (1) The present invention adopts the iterative reweighted minimum covariance determinant method to select the optimal subset from the spatial neighborhood of each spatial grid point to estimate the robust local spectral matrix, which effectively suppresses the interference of outliers in the field survey data on covariance estimation, fundamentally overcomes the defect of traditional principal component analysis being sensitive to abnormal data, and significantly improves the reliability and stability of degradation assessment results.
[0057] (2) The present invention introduces a row sparsity mixed norm penalty term into the unified optimization problem to make the optimal load matrix obtained by the solution automatically generate row sparsity, thereby directly screening out a few key variables that drive degradation; combined with the preset configurable rule engine, it can clearly determine the degradation stress type of each grid point, providing an intuitive and quantifiable scientific basis for precise ecological restoration.
[0058] (3) This invention introduces a spatial smoothing penalty term in the unified optimization problem. By measuring the distance between the principal components of adjacent grid points and the projection matrix of the subspace, and forcing it to be minimized, the spatial continuity constraint of the degradation features is realized. Compared with the grid-by-grid independent analysis method without spatial smoothing, it effectively reduces the fragmentation of the evaluation results, makes the extracted degradation patch boundaries more complete and the spatial consistency stronger, and significantly improves the practical value of the spatial early warning layer.
[0059] (4) The present invention uses the inaccurate alternating direction multiplier method to decompose the large-scale unified optimization problem into parallel subproblems at each grid point, and updates the load matrix, auxiliary variables and dual variables in parallel in each iteration step; this distributed solution strategy can efficiently deploy the computing task on the GPU cluster, and control the computing time of processing hundreds of thousands of grid points in a single process to the order of several hours, which meets the timeliness requirements of monitoring the degradation of multi-star chive populations in large-scale alpine grasslands. Attached Figure Description
[0060] Other features, objects, and advantages of the invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings. The drawings are for illustrative purposes only and are not intended to limit the invention. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings:
[0061] Figure 1 This is a flowchart of a method for classifying and spatially warning about the degradation levels of *Allium chinense* populations in alpine grasslands, provided in an embodiment of the present invention.
[0062] Figure 2 This is an overall technical roadmap of a method for classifying and spatially warning about the degradation levels of *Allium chinense* populations in alpine grasslands, provided by an embodiment of the present invention.
[0063] Figure 3 This is a schematic diagram of the degradation level spatial early warning layer provided in an embodiment of the present invention;
[0064] Figure 4 This is a schematic diagram of the structure of a system for classifying the degradation level of *Allium chinense* populations in alpine grasslands and providing a spatial early warning system, provided in an embodiment of the present invention.
[0065] Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation
[0066] The present invention will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and not intended to limit it. Furthermore, it should be noted that, for ease of description, the accompanying drawings show only the parts relevant to the present invention, and not all of the structures.
[0067] Before discussing the exemplary embodiments in more detail, it should be mentioned that some exemplary embodiments are described as processes or methods depicted as flowcharts. Although the flowcharts describe operations (or steps) as being processed sequentially, many of these operations (or steps) may be performed in parallel, concurrently, or simultaneously. Furthermore, the order of the operations may be rearranged. The process may be terminated when its operation is completed, but may also have additional steps not included in the figures. The process may correspond to a method, function, procedure, subroutine, subroutine, etc.
[0068] Example 1
[0069] like Figure 1 As shown, Embodiment 1 of the present invention provides a method 100 for classifying and spatially warning of degradation levels of *Allium chinense* populations in alpine grasslands. The method 100 specifically includes the following steps:
[0070] Step S110: Divide the target region into multiple spatial grid points and obtain the multi-star degradation-related feature variables of each spatial grid point, and construct the feature vector of each spatial grid point.
[0071] In some embodiments, the target region is divided into N spatial grid points, and for each grid point... Collect p feature variables related to the degradation of multi-star chives to form a feature vector. .
[0072] Among them, the characteristic variables include at least the following species-specific indicators: plant height of *Leekia spicata*, flower diameter, leaf width, bulb density, number of divisions, soil humus layer thickness, soil bulk density, soil moisture content, distance from the trail, and coverage of major competing species.
[0073] Step S120: Define the spatial neighborhood information of each spatial grid point, and obtain the local spectral matrix of each spatial grid point based on the feature vector and spatial neighborhood information of each spatial grid point through the robust covariance estimation method.
[0074] In some embodiments, for each grid point i, its spatial neighborhood set is defined. The neighborhood window is adaptively determined based on the spatial resolution, typically ranging from 3×3 to 7×7.
[0075] Specifically, for the boundary grid points and corner points of the target region, a reflection-filling method is used to expand their spatial neighborhood to the full window size; if the expansion is still insufficient, the sample subset size h is adjusted. , The covariance matrix is selected adaptively between 60% and 80%, and the actual sample size is ensured to be no less than 1.5 times p, so that the covariance matrix can be estimated.
[0076] The Iterative Reweighted Minimum Covariance Determinant (IR-MCD) method is used to find the optimal subset containing h samples from the neighborhood data matrix, and the covariance matrix of this subset is used as the grid. Robust local spectral matrix ,Right now When the neighborhood sample size When this happens, the system automatically marks the output results with a warning indicating low data reliability.
[0077] Step S130: Construct and solve a unified optimization problem based on the local spectral matrix of each spatial grid point, and output the optimal load matrix of each spatial grid point through distributed iterative optimization. The unified optimization problem aims to maximize the projection variance, filter key variables by row sparsity constraints, and ensure the principal component continuity of adjacent grid points by spatial smoothness constraints.
[0078] In some embodiments, calculate the d-dimensional (d≥2, typically 2 or 3) principal component loading matrix for all grid points. Maximize the following expression:
[0079]
[0080] Among them: the first item To maximize the total projection variance and ensure that the principal components retain the maximum amount of degradation information; the second term To implement sparse penalties, among which Forced Many rows are all zero, achieving variable interpretability; the third item... As a spatial smoothness penalty, the projection matrix distance between adjacent lattice points principally constitutes the subspace is measured, where The set of edges for all adjacent grid points; orthogonal constraint. Ensure that the columns of the load matrix are orthogonal. It is a d×d identity matrix; The number of principal components that need to be estimated; : Principal component loading matrix at the i-th grid point, with columns representing the directions of each principal component; The set of all grid-point load matrices; : The robust local spectral matrix of the i-th lattice point; The trace of a matrix; : Weight coefficients of the row sparsity penalty term; :matrix Mix Norm (row sparsity penalty), defined as ; Load matrix The Line 1 Column elements; The calculation formula first calculates the elements of each row. Norm, then sum the norms of all rows; : Weight coefficients of the spatial smoothing penalty term; Summing over all adjacent grid pairs; The Frobenius norm of a matrix. ; The squared Frobenius norm of the difference between the projection matrices of adjacent grid points i and j; : The projection matrix corresponding to the load matrix of the i-th grid point (rank is (a positive semidefinite matrix); The projection matrix corresponding to the load matrix of the j-th grid point (rank is...) (a positive semidefinite matrix); identity matrix; Orthogonal constraint: ensures that the column vectors of the load matrix are mutually orthogonal and of unit length.
[0081] In some embodiments, the uniform optimization problem is decomposed into parallel subproblems at each spatial grid point using the inexact alternating direction multiplier method, and auxiliary variables are introduced. and consensus constraints (Orthogonal constraints are applied synchronously to) ,Right now Construct the augmented Lagrangian function:
[0082]
[0083] in: : The auxiliary variable matrix of the i-th grid point, used for ADMM consensus constraint decomposition; : For auxiliary variables The applied orthogonal constraints, and Maintain consistency; The dual variable (Lagrange multiplier) matrix at the i-th lattice point, used for constraint ; : The augmented Lagrange penalty parameter is adaptively adjusted during the iteration process; :by For the augmented Lagrangian function of the penalty parameters; : Matrix inner product , represents the inner product penalty between the dual variable and the constraint violation quantity; The sum of the linear penalty terms for the dual variables of all lattice points; The sum of quadratic penalty terms for all grid points, forced and Towards consensus; and The squared Frobenius norm of the difference; The i-th grid point is at The projection variance trace on; :matrix Mix Norm (sparse line is penalized). ; The Frobenius norm of the squared difference between the auxiliary variable projection matrices of adjacent grid points i and j acts on... Spatial smoothing penalty; The projection matrix of the auxiliary variable at the i-th grid point (rank is...) (a positive semidefinite matrix); The projection matrix of the auxiliary variable at the j-th grid point (rank is...) (a positive semidefinite matrix); The Frobenius norm of a matrix. .
[0084] For augmented Lagrangian functions The overall structure is explained as follows: In this formula, The negative projection variance term transforms the maximization problem into a minimization problem. To implement sparse penalty terms; For the linear penalty term of the dual variable; This is a secondary penalty item, subject to mandatory consensus constraints. This is a spatial smoothing penalty term that acts on auxiliary variables.
[0085] The following sub-problems are executed alternately:
[0086] (1) - Update subproblem (parallel across all grid points): fixed and The problem is for each Separate into:
[0087] ;
[0088] The proximal gradient method is used for iterative solution. Each step includes: gradient descent step, row sparse soft thresholding proximal operator, and then orthogonality is restored through QR decomposition.
[0089] in, : The principal component loading matrix to be updated at the i-th grid point; : The auxiliary variable of the i-th grid point (currently fixed value); : The dual variable of the i-th grid point (currently fixed value); The set of all grid-point auxiliary variables (fixed); : The set of all lattice dual variables (fixed); Grid The projected variance trace; :matrix Mix Norm, ; : Augmenting Lagrange near-end target points; : Second-order proximal term, measuring Distance to the nearest target point;
[0090] - Update subproblem (space smoothing step): fixed The update must consider both proximal terms and spatial smoothing terms. Since a closed-form solution is unavailable, this invention employs a heuristic two-step approximation:
[0091] Step 1 (Relaxation and Smoothing of Projection Matrix): Let (This can be viewed as a relaxation of the projection matrix), solve the graph Laplace smoothing problem: ,in Let be the near-end target point. This problem is a convex quadratic problem, which can be solved efficiently using the conjugate gradient method to obtain a smoothed matrix. ;
[0092] Step 2 (Truncation of SVD to restore orthogonal basis): For each Perform truncated singular value decomposition, retaining only the first d largest eigenvalues and their corresponding eigenvectors to form the updated eigenvalues. directly satisfy According to the Eckart-Young theorem, this projection is the nearest rank under the Frobenius norm. Positive semidefinite matrix.
[0093] in, Near-end target points, integrated and dual variable information; The current relaxed representation of the projection matrix (rank d positive semi-definite) to be updated; The projection matrix after Tulaplace smoothing (may no longer satisfy the rank requirement) constraint); The set of all smooth projection matrices; : Projection matrix form of near-end target points; : Data fidelity term, which makes the smoothed matrix Approaching the near target point; : Graph smoothing penalty term, coefficient Simplification by merging the spatial smoothing term and the proximal term; The set of edges between adjacent grid points. ; The smoothed matrix is symmetric positive semi-definite but does not satisfy the rank requirement. constraint.
[0094] Updated The orthogonal matrix formed by the first d eigenvectors automatically satisfies .
[0095] The strategy showed stable convergence in the experiment, but did not guarantee a global optimal solution.
[0096] (3) Dual variable update: The conditional residual after iteration to KKT is reduced to When the order of magnitude reaches or the preset maximum number of iterations is reached, the initialization adopts a spectral initialization strategy: perform ordinary PCA on each grid point individually to obtain the initial value. This method employs an inexact ADMM iterative framework, which empirically achieves stable convergence, although the specific convergence behavior depends on parameter settings and data characteristics.
[0097] in, : The dual variable of the i-th lattice point; Standard ADMM dual update, Consensus constraints on the amount of violations; : Initial load matrix, obtained by single-grid PCA (spectral initialization).
[0098] Step S140: Calculate the degradation intensity index based on the optimal load matrix and local spectrum matrix of each spatial grid point, identify the dominant degradation variable based on the row comprehensive load of the optimal load matrix, and determine the degradation type of each spatial grid point through a configurable rule engine.
[0099] In some embodiments, this step specifically includes:
[0100] (1) Degradation intensity index calculation: based on the converged optimal load matrix and robust local spectral matrix Calculate the degradation intensity index for each grid point. The calculation formula is: ; in, represents the degradation intensity index of the i-th grid point, which measures the total projection variance of the grid point on the principal component subset space. The larger the value, the higher the degree of degradation.
[0101] (2) Identification of dominant degradation variables: Calculation of the optimal load matrix Each line The norm (i.e., the combined load value) is calculated using the following formula: , ; in, This represents the combined loading value of the k-th variable among all d principal components; For the optimal load matrix The element in the k-th row and l-th column; : Variable index; Calculate the combined load value for each variable. and the obtained comprehensive load values Sort the variables from largest to smallest, and select one or more of the top-ranked variables as the dominant degenerate variables for that grid point.
[0102] (3) Dominant Degradation Type Identification: The system matches degradation types through a rule engine. This engine has several preset degradation types and their corresponding dominant degradation variable condition combinations. Rules can be constructed based on expert experience or data-driven methods and can be optimized and adjusted through historical accumulated data. By matching the identified dominant degradation variable with the conditions in the rule engine, the degradation type of the grid point can be determined. As an example, the preset degradation types are shown in Table 1:
[0103] Table 1
[0104]
[0105] Step S150: Perform spatial segmentation and connectivity analysis on the degradation intensity index, extract continuous degradation patches and classify degradation levels, and output a spatial early warning layer containing patch boundaries, degradation levels and degradation types.
[0106] In some embodiments, the degradation intensity index The global distribution is analyzed for distribution characteristics: if it shows a significant bimodal distribution, the Otsu adaptive threshold is used; if it shows a unimodal distribution or a gradual change, a threshold based on the interquartile range is used. If the study area is large or has significant topographical differences, it can be divided into several sub-regions for separate processing; among them, : A set of degradation intensity indices for all grid points; Otsu adaptive thresholding: An automatic thresholding method based on the bimodality of gray-level histograms, maximizing inter-class variance; The upper quartile (75th percentile) of the data; Interquartile range The lower quartile (25th percentile); Anomaly detection thresholds based on interquartile ranges are often used to identify extreme values that significantly deviate from the mainstream distribution.
[0107] Eight-neighbor spatial connectivity analysis was performed on the binarization results;
[0108] Calculate the patch density index Landscape pattern indicators; patch density index This value is used to measure the degree of landscape fragmentation; a larger value indicates that the degraded areas are more dispersed and fragmented. Based on patch area and average degradation intensity The system performs grading and outputs a spatial warning layer containing patch boundaries, degradation levels, and dominant degradation types, including the average degradation intensity. The calculation formula is: ; Where: c: patch index; The set of grid points contained in the c-th degraded plaque; plaque The number of grid points in the data; Grid The degradation intensity index.
[0109] Preferably, based on patch area and average degradation intensity The specific method for classifying the grading is as follows: based on the preset area threshold and degradation intensity threshold, the patches are divided into three levels: mild degradation, moderate degradation, and severe degradation.
[0110] For example, patches with an area smaller than a preset area threshold (e.g., 0.1 hectares) are considered sporadic degradation points and are not classified separately; patches with an area greater than or equal to the preset area threshold are classified according to the numerical range of their average degradation intensity: those below the lower quartile Q1 are mildly degraded, those between the lower quartile Q1 and the upper quartile Q3 are moderately degraded, and those above the upper quartile Q3 are severely degraded.
[0111] In this study, patch boundaries are obtained directly through connected component analysis; the dominant degradation type of a patch is determined by a majority vote of the degradation types of all grid points within the patch. The final output is a spatial warning layer containing patch boundary coordinates, degradation levels, and dominant degradation types.
[0112] Based on the above embodiments, the core beneficial effects of the present invention are as follows:
[0113] (1) The present invention adopts the iterative reweighted minimum covariance determinant method to select the optimal subset from the spatial neighborhood of each grid point to estimate the robust local spectral matrix, which effectively suppresses the interference of outliers in the field survey data on covariance estimation, fundamentally overcomes the defect of traditional principal component analysis being sensitive to abnormal data, and significantly improves the reliability and stability of degradation assessment results.
[0114] (2) The present invention introduces a row sparsity mixed norm penalty term into the unified optimization problem to make the optimal load matrix obtained by the solution automatically generate row sparsity, thereby directly screening out a few key variables that drive degradation; combined with the preset configurable rule engine, it can clearly determine the degradation stress type of each grid point, providing an intuitive and quantifiable scientific basis for precise ecological restoration.
[0115] (3) This invention introduces a spatial smoothing penalty term in the unified optimization problem. By measuring the distance between the principal components of adjacent grid points and the projection matrix of the subspace, and forcing it to be minimized, the spatial continuity constraint of the degradation features is realized. Compared with the grid-by-grid independent analysis method without spatial smoothing, it effectively reduces the fragmentation of the evaluation results, makes the extracted degradation patch boundaries more complete and the spatial consistency stronger, and significantly improves the practical value of the spatial early warning layer.
[0116] (4) The present invention uses the inaccurate alternating direction multiplier method to decompose the large-scale unified optimization problem into parallel subproblems at each grid point, and updates the load matrix, auxiliary variables and dual variables in parallel in each iteration step; this distributed solution strategy can efficiently deploy the computing task on the GPU cluster, and control the computing time of processing hundreds of thousands of grid points in a single process to the order of several hours, which meets the timeliness requirements of monitoring the degradation of multi-star chive populations in large-scale alpine grasslands.
[0117] Example 2
[0118] This example focuses on the Jiucaiping Scenic Area in Bijie, Guizhou, approximately 5 km away. 2 The core degradation monitoring area was the experimental area, employing a 5 m × 5 m spatial resolution to generate approximately 200,000 spatial grid points. The data for each grid point was obtained through field quadrat surveys, UAV multispectral remote sensing, and digital elevation models. =11 indicators: plant height of *Leekia serratifolia*, flower diameter, leaf width, bulb density, division rate in the current year, soil humus layer thickness, soil bulk density, soil moisture content, distance from the trail, coverage of competing species, and slope.
[0119] A 5×5 neighborhood window was used for robust covariance estimation of IR-MCD, with h selected at approximately 75% of the sample size. Reflection-filled expansion of the neighborhood was applied to the boundary grid points, and the neighborhood sample size was... The grid points are marked with confidence warnings in the output. This step is performed on a 32-core CPU.
[0120] The value of d was set to 2. A total of 386 field plots were established within the experimental area. Two ecological experts independently assessed the degradation level based on the characteristics of the *Allium chinense* population, with a Kappa coefficient of 0.82 for consistency. Using these plots as the validation set, hyperparameters were determined through five-fold cross-validation. , At this point, the overall accuracy of the quadrat-level classification is OA=88.2%.
[0121] Implement distributed ADMM solution using PyTorch on 4 GPUs (single machine). The update step is processed in parallel on each GPU in a grid-like manner; The update step uses a heuristic two-step approximation. Under GPU acceleration, the overall computation time is on the order of several hours, depending on the hardware configuration.
[0122] Degradation intensity The variance contribution rate of the first principal component reached .high The load norms of the region were mainly concentrated on soil moisture content, distance from the trail, and bulb density. Based on rule engine matching, the highly degraded areas were classified into two types: human trampling-soil drought stress and population density competition.
[0123] The distribution exhibits a significant bimodal characteristic, and highly degraded areas were extracted using the Otsu method. Eight-neighbor connectivity analysis identified several degraded hotspot patches with an area >0.1 hectares. Compared to the method without spatial smoothing (… Using the grid-by-grid independent analysis method as a benchmark, the patch fragmentation index of the method in this invention shows a decreasing trend, and the spatial continuity and patch integrity are significantly improved. The final output is a spatial warning layer, indicating the boundary, level, and dominant degradation mechanism of each patch.
[0124] Example 3
[0125] Figure 4 This is a schematic diagram of the structure of a high-altitude grassland ivy population degradation level classification and spatial early warning system 400 provided in Embodiment 3 of the present invention, as shown in the figure. Figure 4 As shown, the system includes:
[0126] The data acquisition and gridding module 410 is used to divide the target area into multiple spatial grids and obtain the multi-star degradation-related feature variables of each spatial grid, and construct the feature vector of each spatial grid.
[0127] The local spectral matrix estimation module 420 is used to define the spatial neighborhood information of each spatial grid point, and obtain the local spectral matrix of each spatial grid point through a robust covariance estimation method based on the feature vector of each spatial grid point and its spatial neighborhood information.
[0128] The unified optimization problem solving module 430 is used to construct and solve a unified optimization problem based on the local spectral matrix of each spatial grid point. It outputs the optimal load matrix of each spatial grid point through distributed iterative optimization. The unified optimization problem aims to maximize the projection variance, filter key variables by row sparsity constraints, and ensure the principal component continuity of adjacent grid points by spatial smoothness constraints.
[0129] The degradation type discrimination module 440 is used to calculate the degradation intensity index based on the optimal load matrix and local spectrum matrix of each spatial grid point, identify the dominant degradation variable based on the row comprehensive load of the optimal load matrix, and determine the degradation type of each spatial grid point through a configurable rule engine.
[0130] The spatial early warning output module 450 is used to perform spatial segmentation and connectivity analysis on the degradation intensity index, extract continuous degradation patches and classify degradation levels, and output a spatial early warning layer containing patch boundaries, degradation levels and degradation types.
[0131] The alpine meadow *Allium chinense* population degradation level classification and spatial early warning system provided in this embodiment of the invention can execute the alpine meadow *Allium chinense* population degradation level classification and spatial early warning method provided in any of the above embodiments of the invention. It has the corresponding functions and beneficial effects of executing the alpine meadow *Allium chinense* population degradation level classification and spatial early warning method. For detailed process, please refer to the relevant operations of the alpine meadow *Allium chinense* population degradation level classification and spatial early warning method in the foregoing embodiments.
[0132] Example 4
[0133] Figure 5 This is a schematic diagram of the structure of an electronic device provided in Embodiment 4 of the present invention. The electronic device 10 is intended to represent various forms of digital computers, and may also represent various forms of mobile devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the invention described and / or claimed herein.
[0134] like Figure 5As shown, the electronic device 10 includes at least one processor 11 and a memory, such as a read-only memory (ROM) 12 or a random access memory (RAM) 13, communicatively connected to the at least one processor 11. The memory stores computer programs executable by the at least one processor. The processor 11 can perform various appropriate actions and processes based on the computer program stored in the ROM 12 or loaded into the RAM 13 from storage unit 18. The RAM 13 can also store various programs and data required for the operation of the electronic device 10. The processor 11, ROM 12, and RAM 13 are interconnected via a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.
[0135] Multiple components in electronic device 10 are connected to I / O interface 15, including: input unit 16, such as keyboard, mouse, etc.; output unit 17, such as various types of displays, speakers, etc.; storage unit 18, such as disk, optical disk, etc.; and communication unit 19, such as network card, modem, wireless transceiver, etc. Communication unit 19 allows electronic device 10 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.
[0136] Processor 11 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of processor 11 include, but are not limited to, central processing unit (CPU), graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, digital signal processors (DSPs), and any suitable processor, controller, microcontroller, etc. Processor 11 executes the alpine grassland multi-star chive population degradation grading and spatial early warning method described above.
[0137] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this invention can be achieved, and this is not limited herein.
[0138] The above embodiments are merely illustrative examples and are not intended to limit the implementation. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively list all possible implementations. However, obvious variations or modifications derived therefrom are still within the scope of protection of this invention.
Claims
1. A method for classifying and spatially warning the degradation grades of Allium polyrhizum populations in alpine meadow, characterized in that, include: The target region is divided into multiple spatial grid points, and the multi-star degradation-related feature variables of each spatial grid point are obtained to construct the feature vector of each spatial grid point. Define the spatial neighborhood information of each spatial grid point, and obtain the local spectral matrix of each spatial grid point through the robust covariance estimation method based on the feature vector of each spatial grid point and its spatial neighborhood information. A unified optimization problem is constructed and solved based on the local spectral matrix of each spatial grid point. The optimal load matrix of each spatial grid point is output through distributed iterative optimization. The unified optimization problem aims to maximize the projection variance, filter key variables by row sparsity constraints, and ensure the principal component continuity of adjacent grid points by spatial smoothness constraints. The degradation intensity index is calculated based on the optimal load matrix and local spectrum matrix of each spatial grid point, and the dominant degradation variable is identified based on the row comprehensive load of the optimal load matrix. The degradation type of each spatial grid point is determined by a configurable rule engine. Spatial segmentation and connectivity analysis are performed on the degradation intensity index to extract continuous degradation patches and classify degradation levels, outputting a spatial early warning layer containing patch boundaries, degradation levels, and degradation types.
2. The alpine meadow Allium polyrhizum population degradation grade classification and spatial early warning method according to claim 1, characterized in that, Based on the eigenvectors and spatial neighborhood information of each spatial grid point, the local spectral matrix of each spatial grid point is obtained through a robust covariance estimation method, including: The iterative reweighted minimum covariance determinant method is used to select the optimal subset from the spatial neighborhood of each spatial grid point, where the sample size reaches a preset proportion threshold. The covariance matrix of the optimal subset is used as the local spectral matrix of the spatial grid point.
3. The method for classifying and spatially warning of degradation levels of *Allium chinense* populations in alpine grasslands according to claim 1, characterized in that, When defining the spatial neighborhood information of each spatial grid point, for spatial grid points located at the boundary of the target region, a reflection-filling method is used to expand their spatial neighborhood to the full window size; When the actual sample size of the expanded spatial neighborhood is lower than the preset sample size threshold, a data credibility warning message is marked in the output result.
4. The method for classifying and spatially warning of degradation levels of *Allium chinense* populations in alpine grasslands according to claim 1, characterized in that, A unified optimization problem is constructed and solved based on the local spectral matrices of each spatial grid point. The optimal load matrix of each spatial grid point is output through distributed iterative optimization, including: Initialize the load matrix of each spatial grid point, decompose the unified optimization problem into parallel subproblems of each spatial grid point using the inexact alternating direction multiplier method, and introduce auxiliary variables and dual variables; The parallel subproblems of each spatial grid point are solved iteratively. In each iteration, the load matrix, auxiliary variables and dual variables of each spatial grid point are updated in parallel until the preset convergence condition is met. The converged load matrix is then output as the optimal load matrix. The row sparsity constraint is achieved by applying a mixture norm penalty to the load matrix, and the spatial smoothness constraint is achieved by measuring the distance between the principal components of the subspace projection matrix of adjacent grid points.
5. The method for classifying and spatially warning of degradation levels of *Allium chinense* populations in alpine grasslands according to claim 4, characterized in that, Parallel updates of auxiliary variables for each spatial grid point include: The combination of the load matrix and dual variables in the current iteration step is taken as the near-end target point, and its projection matrix is calculated. The projection matrix is smoothed using a graph Laplacian smoothing method to obtain a smoothed projection matrix. The smoothed projection matrix is subjected to truncated singular value decomposition to obtain eigenvalues, which are then sorted. The updated auxiliary variables are constructed by taking the eigenvectors corresponding to the n largest eigenvalues in the eigenvalue sequence, where n is the preset principal component dimension.
6. The method for classifying and spatially warning of degradation levels of *Allium chinense* populations in alpine grasslands according to claim 1, characterized in that, The degradation intensity index is calculated based on the optimal load matrix and local spectral matrix of each spatial grid point, including: Project the optimal load matrix of each spatial grid point onto the local spectral matrix of that spatial grid point, and calculate the projection variance trace of each spatial grid point. The projection variance trace of each spatial grid point is used as the degradation intensity index of the corresponding spatial grid point.
7. The method for classifying and spatially warning of degradation levels of *Allium chinense* populations in alpine grasslands according to claim 6, characterized in that, Based on the optimal load matrix, the row-synthesized load identifies the dominant degradation variable, and a configurable rule engine determines the degradation type of each spatial grid point, including: Calculate the row norm of each row of the optimal load matrix for each spatial grid point, use the row norm as the composite load value of that row, and identify one or more variables with the largest composite load value as the dominant degenerate variables of that spatial grid point. Based on the identified dominant degradation variable, the corresponding degradation type is matched using the configurable rule engine; The rule engine has several pre-defined degradation types and their corresponding dominant degradation variable condition combinations.
8. The method for classifying and spatially warning of degradation levels of *Allium chinense* populations in alpine grasslands according to claim 1, characterized in that, Spatial segmentation and connectivity analysis are performed on the degradation intensity index to extract continuous degradation patches and classify degradation levels. A spatial warning layer containing patch boundaries, degradation levels, and degradation types is output, including: A global distribution analysis of the degradation intensity index of each spatial grid point is performed, and a segmentation threshold is adaptively selected for binarization segmentation based on the distribution characteristics. Connectivity analysis was performed on the binarized segmentation results to extract continuous spatially connected regions as degenerate patches. Calculate the average degradation intensity and patch area for each degraded patch, and classify the degradation level of each spatial grid point based on the average degradation intensity and patch area; Integrate the degradation types of the grid points corresponding to each degradation patch, and output a spatial warning layer that includes patch boundaries, degradation levels, and degradation types.
9. The method for classifying and spatially warning of degradation levels of *Allium chinense* populations in alpine grasslands according to claim 8, characterized in that, The step of adaptively selecting a segmentation threshold based on distribution characteristics for binarization segmentation includes: A bimodality test was performed on the global distribution of the degradation intensity index; If the global distribution exhibits a significant bimodal distribution, the segmentation threshold is determined using the method of maximizing inter-class variance; otherwise, a statistical threshold based on the interquartile range is used for segmentation.
10. A system for classifying and spatially warning of degradation levels of *Allium chinense* populations in alpine grasslands, characterized in that... The system is used to implement the method for classifying and spatially warning of degradation levels of *Allium chinense* populations in alpine grasslands according to any one of claims 1 to 9, the system comprising: The data acquisition and gridding module is used to divide the target area into multiple spatial grids and obtain the multi-star degradation-related feature variables of each spatial grid, and construct the feature vector of each spatial grid. The local spectral matrix estimation module is used to define the spatial neighborhood information of each spatial grid point. Based on the feature vectors of each spatial grid point and its spatial neighborhood information, the local spectral matrix of each spatial grid point is obtained through a robust covariance estimation method. The unified optimization problem solving module is used to construct and solve the unified optimization problem based on the local spectral matrix of each spatial grid point. It outputs the optimal load matrix of each spatial grid point through distributed iterative optimization. The unified optimization problem aims to maximize the projection variance, filter key variables by row sparsity constraints, and ensure the principal component continuity of adjacent grid points by spatial smoothness constraints. The degradation type discrimination module is used to calculate the degradation intensity index based on the optimal load matrix and local spectrum matrix of each spatial grid point, identify the dominant degradation variable based on the row comprehensive load of the optimal load matrix, and determine the degradation type of each spatial grid point through a configurable rule engine. The spatial early warning output module is used to perform spatial segmentation and connectivity analysis on the degradation intensity index, extract continuous degradation patches and classify degradation levels, and output a spatial early warning layer containing patch boundaries, degradation levels and degradation types.