Typical watershed vegetation measure intelligent extraction and function evaluation method driven by multi-source remote sensing

Abstract

The invention discloses a multi-source remote sensing driven typical watershed vegetation measure intelligent extraction and function evaluation method, particularly relates to the technical field of ecological remote sensing and water and soil conservation, and is used for solving the problems that in the prior art, vegetation measure extraction precision is insufficient, and function evaluation and dynamic optimization are disjointed. Vegetation coverage information is extracted by fusing multispectral and high-resolution remote sensing data, and a spatial correlation network is constructed by combining terrain and hydrological factors; quantizing a vegetation measure function index by utilizing topological analysis, and judging a water and soil conservation demand through a dynamic threshold value; a causal gradient field and a neural information bottleneck compression technology are introduced to identify and optimize a key region, and a vegetation adjustment suggestion is generated based on spatial network feedback, so that a full-chain technical breakthrough from data fusion, function evaluation to closed-loop optimization is realized, and accurate management and ecological benefits of vegetation measures in a complex watershed environment are remarkably improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of ecological remote sensing and soil and water conservation, and more particularly, relates to a multi-source remote sensing driven typical watershed vegetation measure intelligent extraction and function evaluation method. BACKGROUND

[0002] With the continuous development of remote sensing technology, multi-source remote sensing data has been widely applied in the fields of vegetation coverage monitoring, soil and water conservation measure evaluation, and soil erosion analysis. However, the existing remote sensing extraction methods still have certain limitations in the intelligent extraction and function evaluation of typical watershed vegetation measures, making it difficult to meet the efficient and accurate needs.

[0003] After searching, a kind of rapid remote sensing extraction method of ginger based on typical phenology and film net characteristics (CN114332628A) is disclosed. This patent realizes the dynamic monitoring of ginger area, yield and spatial distribution by combining the phenological characteristics and cultivation management measures (such as the use of white mulch and black shading net) in the growth process of ginger. However, this technical solution mainly focuses on the extraction of a single crop (ginger), and lacks comprehensive extraction capability for multiple vegetation measures. In addition, its method relies on specific mulch and shading net characteristics, which has a limited scope of application and is difficult to be applied to vegetation measure extraction in other vegetation types or complex watershed environments.

[0004] After searching, an evaluation method for soil erosion and related equipment (CN118212520A) is disclosed. This patent realizes the accurate evaluation of soil erosion amount by extracting remote sensing image information of the study area, combining seasonal changes of rainfall and vegetation coverage, and slope factors. However, this technical solution focuses on soil erosion evaluation and fails to fully consider the function evaluation of vegetation measures and their synergistic effect with soil and water conservation. At the same time, this method has high dependence on training sample set, which may lead to a decrease in the accuracy of evaluation results in the absence of sufficient samples. In addition, its method does not explicitly involve the fusion processing of multi-source remote sensing data, limiting its application potential in complex watershed environments.

[0005] The above problems show that the existing remote sensing extraction and evaluation methods still have deficiencies in multi-source data fusion, comprehensive extraction of multiple vegetation measures, function evaluation, and adaptability to complex watershed environments, resulting in defects in the intelligent extraction of multiple vegetation measures. SUMMARY

[0006] In order to overcome the above-mentioned defects of the prior art, the embodiments of the present application provide a multi-source remote sensing driven typical watershed vegetation measure intelligent extraction and function evaluation method to solve the problems raised in the background art.

[0007] To achieve the above-mentioned purpose, the present application provides the following technical solutions:

[0008] A method for intelligent extraction and functional evaluation of vegetation measures in typical watersheds driven by multi-source remote sensing includes the following steps:

[0009] S1. Acquire multi-source remote sensing image data of the study area, extract vegetation cover information based on spectral and texture features, and perform multi-level segmentation of vegetation cover information to generate multi-scale vegetation distribution maps.

[0010] S2. By analyzing the spatial structure characteristics of multi-scale vegetation distribution maps, a spatial correlation network of vegetation measures is constructed by combining topographic and hydrological factors.

[0011] S3. Perform topological analysis on the spatial network and calculate the functional evaluation index of vegetation measures by extracting node degree centrality, betweenness centrality and clustering coefficient.

[0012] S4. Compare the functional evaluation index with the preset benchmark threshold to determine whether the vegetation measures meet the soil and water conservation requirements.

[0013] S5. When the soil and water conservation requirements are not met, construct the causal gradient field of the soil and water conservation functional potential field, extract the mutation points of nodes in the gradient flow through neural information bottleneck compression, and define the priority optimization area.

[0014] S6. Based on the spatial location and functional requirements of the priority optimization area, generate suggestions for adjusting vegetation measures and update the spatial association network.

[0015] In a preferred embodiment, multi-source remote sensing image data of the study area is acquired, vegetation cover information is extracted based on spectral and texture features, and the vegetation cover information is segmented at multiple levels to generate a multi-scale vegetation distribution map, including:

[0016] Acquire multispectral and high-resolution panchromatic remote sensing images of the study area, perform radiometric and geometric corrections on the multispectral remote sensing images, and generate preprocessed multi-source remote sensing image data.

[0017] Based on the preprocessed multispectral remote sensing images, the normalized vegetation index and enhanced vegetation index were calculated, and the spectral characteristics of the vegetation-covered area were extracted.

[0018] Based on the preprocessed high-resolution panchromatic remote sensing image, the texture energy value and contrast of the vegetation-covered area are extracted using the gray-level co-occurrence matrix algorithm to generate a texture feature set.

[0019] The spectral and texture feature sets are input into a multi-scale segmentation algorithm. By iteratively merging adjacent pixel blocks, vegetation cover patches of different spatial scales are generated, forming a multi-scale vegetation distribution map.

[0020] In a preferred embodiment, by analyzing the spatial structure characteristics of multi-scale vegetation distribution maps and combining topographic and hydrological factors, a spatial correlation network of vegetation measures is constructed, including:

[0021] Based on multi-scale vegetation distribution maps, the aggregation index and proximity index of vegetation cover patches are extracted to generate a spatial structure feature set.

[0022] Obtain digital elevation model data of the study area, calculate slope, terrain curvature and elevation variation coefficient, and generate a set of terrain factors;

[0023] Data on cumulative runoff and river network density in the study area were obtained, and the runoff path of the watershed was simulated based on the hydrological model to generate a set of hydrological factors.

[0024] By inputting the spatial structure feature set, topographic factor set, and hydrological factor set into the spatial network model, and by coupling the spatial adjacency relationship of vegetation patches with the topographic-hydrological interaction weight, a spatial association network of vegetation measures is constructed.

[0025] In a preferred embodiment, topological analysis is performed on the spatially associated network, and a functional evaluation index of vegetation measures is calculated by extracting node degree centrality, betweenness centrality, and clustering coefficients, including:

[0026] Based on the spatial association network, the degree centrality of each node is calculated. The degree centrality is the ratio of the number of edges directly connected to the node to the maximum possible number of edges connected to the network.

[0027] Calculate the betweenness centrality of each node, which is the ratio of the number of times the node is on the shortest path among other nodes to the total number of shortest paths in the network;

[0028] Calculate the clustering coefficient for each node. The clustering coefficient is the ratio of the actual number of connecting edges between the node's neighboring nodes to the theoretical maximum number of connecting edges.

[0029] By inputting degree centrality, betweenness centrality, and clustering coefficient into the functional assessment model, a functional assessment index for vegetation measures is generated.

[0030] The functional evaluation index is normalized, and the normalized value range is from 0 to 1.

[0031] In a preferred embodiment, the functional evaluation model is calculated as the product of degree centrality and betweenness centrality plus the interaction term between the clustering coefficient and degree centrality.

[0032] In a preferred embodiment, the functional assessment index is compared with a preset benchmark threshold to determine whether the vegetation measures meet the soil and water conservation requirements, including:

[0033] Obtain preset benchmark thresholds, which include dynamic quantile thresholds based on historical soil and water conservation data and empirical thresholds from an expert knowledge base;

[0034] The normalized functional assessment index is compared with the dynamic quantile threshold. If the normalized functional assessment index is lower than the 25th percentile of the dynamic quantile threshold, it is determined that the soil and water conservation requirements are not met.

[0035] The normalized functional assessment index is compared with the empirical threshold. If the index value is lower than the lower limit of the empirical threshold and the index value in a local area has decreased continuously for more than three cycles, it is determined that the soil and water conservation requirements are not met.

[0036] Based on the global and local judgment results, a state label for meeting the soil and water conservation requirements is generated. The state label includes three categories: met, critical, and not met.

[0037] In a preferred embodiment, when the soil and water conservation requirements are not met, a causal gradient field of the soil and water conservation functional potential field is constructed. The abrupt change points of nodes in the gradient flow are extracted through neural information bottleneck compression to define the priority optimization region, including:

[0038] A potential energy field for soil and water conservation functions is constructed. This potential energy field is generated by the product of the normalized functional assessment index and the topographic-hydrological interaction weight.

[0039] Based on the potential energy field of soil and water conservation function, the causal gradient direction and intensity of each node in the potential energy field are calculated by the gradient descent algorithm to form a causal gradient field.

[0040] The causal gradient field is input into the neural information bottleneck compression model, which extracts the mutation points in the gradient flow through a multilayer perceptron.

[0041] Based on the spatial distribution density and intensity threshold of mutation points, a priority optimization region is defined. The priority optimization region is the region where the mutation point density is higher than the preset density threshold and the average intensity exceeds the preset intensity threshold.

[0042] In a preferred embodiment, the criteria for determining a mutation point are that the gradient direction mutation angle exceeds a preset angle threshold and the gradient intensity change rate is significantly higher than the historical average level.

[0043] In a preferred embodiment, based on the spatial location and functional requirements of the priority optimization area, vegetation measure adjustment suggestions are generated and the spatial association network is updated, including:

[0044] Based on the spatial location of the priority optimization area, outlier values ​​of vegetation cover type and topographic-hydrological interaction weights within the priority optimization area are extracted.

[0045] Based on the outlier values ​​of vegetation cover type and topographic-hydrological interaction weight, suggestions for adjusting vegetation measures are generated. These suggestions include vegetation type replacement, planting density adjustment, and supplementation of soil and water conservation projects.

[0046] Based on the adjustment recommendations, the topographic-hydrological interaction weights in the spatial association network are updated;

[0047] The updated spatial association network is overlaid with the multi-scale vegetation distribution map to generate an optimized vegetation measure distribution map.

[0048] In a preferred embodiment, the weight update rule is to reduce the weight of abnormal nodes and enhance the connection strength of optimized nodes.

[0049] Compared with the prior art, the present invention has the following beneficial effects:

[0050] This invention achieves intelligent extraction and dynamic functional assessment of watershed vegetation measures through multi-source remote sensing data fusion and spatial network modeling. First, multi-level segmentation based on spectral and texture features and the generation of multi-scale vegetation distribution maps significantly improve the extraction accuracy of vegetation cover information in complex environments, overcoming the limitations of traditional methods that rely on single features or specific crop parameters. A spatial correlation network constructed by combining topographic and hydrological factors effectively quantifies the interaction between vegetation measures and the geographical environment, providing multi-dimensional data support for functional assessment. Second, through topological analysis and a dynamic optimization mechanism based on causal gradient fields, accurate assessment of vegetation measure functions and rapid location of key problem areas are achieved. The combination of functional assessment indices and dynamic threshold determination not only avoids the defect of excessive sample dependence but also enhances the model's applicability in data-scarce areas. Furthermore, the optimized area definition and closed-loop adjustment suggestions based on gradient flow mutation points can specifically enhance the synergistic effect of soil and water conservation measures, forming a sustainable ecological governance decision support capability. Through multi-technology coupling and dynamic feedback mechanisms, an efficient and reliable technical path is provided for watershed-scale vegetation measure planning. Attached Figure Description

[0051] Figure 1 This is a flowchart of a method for intelligent extraction and functional evaluation of vegetation measures in a typical watershed driven by multi-source remote sensing, according to the present invention. Detailed Implementation

[0052] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0053] Example:Figure 1 This invention presents a method for intelligent extraction and functional evaluation of vegetation measures in typical watersheds driven by multi-source remote sensing, which includes the following steps:

[0054] S1. Acquire multi-source remote sensing image data of the study area, extract vegetation cover information based on spectral and texture features, and perform multi-level segmentation of vegetation cover information to generate multi-scale vegetation distribution maps.

[0055] S2. By analyzing the spatial structure characteristics of multi-scale vegetation distribution maps, a spatial correlation network of vegetation measures is constructed by combining topographic and hydrological factors.

[0056] S3. Perform topological analysis on the spatial network and calculate the functional evaluation index of vegetation measures by extracting node degree centrality, betweenness centrality and clustering coefficient.

[0057] S4. Compare the functional evaluation index with the preset benchmark threshold to determine whether the vegetation measures meet the soil and water conservation requirements.

[0058] S5. When the soil and water conservation requirements are not met, construct the causal gradient field of the soil and water conservation functional potential field, extract the mutation points of nodes in the gradient flow through neural information bottleneck compression, and define the priority optimization area.

[0059] S6. Based on the spatial location and functional requirements of the priority optimization area, generate suggestions for adjusting vegetation measures and update the spatial association network.

[0060] S1. Acquire multi-source remote sensing image data of the study area, extract vegetation cover information based on spectral and texture features, and perform multi-level segmentation of the vegetation cover information to generate a multi-scale vegetation distribution map, including:

[0061] Acquire multispectral and high-resolution panchromatic remote sensing images of the study area, perform radiometric and geometric corrections on the multispectral remote sensing images, and generate preprocessed multi-source remote sensing image data.

[0062] Based on the preprocessed multispectral remote sensing images, the normalized vegetation index and enhanced vegetation index were calculated, and the spectral characteristics of the vegetation-covered area were extracted.

[0063] Based on the preprocessed high-resolution panchromatic remote sensing image, the texture energy value and contrast of the vegetation-covered area are extracted using the gray-level co-occurrence matrix algorithm to generate a texture feature set.

[0064] The spectral and texture feature sets are input into a multi-scale segmentation algorithm. By iteratively merging adjacent pixel blocks, vegetation cover patches of different spatial scales are generated, forming a multi-scale vegetation distribution map.

[0065] Multispectral and high-resolution panchromatic remote sensing images of the study area were acquired. The multispectral images included visible, near-infrared, and short-wave infrared bands, with a spatial resolution of at least 30 meters, while the high-resolution panchromatic images had a spatial resolution of at least 2 meters. Radiometric and geometric corrections were performed on the multispectral images. Radiometric correction employed a 6S atmospheric correction model to eliminate the effects of atmospheric scattering and solar elevation angle. Geometric correction involved selecting at least 20 uniformly distributed ground control points within the study area and performing quadratic polynomial fitting to generate preprocessed multi-source remote sensing image data. Ground control points were obtained by matching measured GPS coordinates of the study area with corresponding points in the images. The registration error of the geometrically corrected images was controlled within 0.5 pixels.

[0066] Based on preprocessed multispectral remote sensing imagery, the Normalized Difference Vegetation Index (NDI) and Enhanced Vegetation Index (EDI) were calculated. The NDI was calculated by dividing the difference between near-infrared and red reflectance by their sum. The EDI incorporated blue reflectance as an atmospheric correction term, with empirical coefficients set to C1 = 6.0, C2 = 7.5, gain coefficient G = 2.5, and background adjustment parameter L = 1.0. A threshold for vegetation coverage was set; for example, areas with both a NDI greater than 0.3 and an EDI greater than 0.2 were considered vegetation-covered areas, and the spectral characteristics of these areas were extracted.

[0067] Based on preprocessed high-resolution panchromatic remote sensing imagery, a gray-level co-occurrence matrix (GLCM) algorithm was used to extract texture energy values ​​and contrast values ​​for vegetation-covered areas. The calculation window for the GLCM was set to 15×15 pixels, with a step size of 5 pixels and 64 gray levels. The directions included 0°, 45°, 90°, and 135°, and the average of these four directions was taken as the texture energy value and contrast value. The texture energy value was calculated as the sum of the squares of each element in the GLCM, and the contrast value was calculated as the weighted sum of the squares of the differences between each element and its corresponding row and column. These steps generated a texture feature set, which included the texture energy value matrix and contrast matrix for the vegetation-covered areas.

[0068] The texture energy value is calculated as follows:

[0069]

[0070] Where E represents the texture energy value, P(i,j) represents the element in the i-th row and j-th column of the gray-level co-occurrence matrix, i and j represent the row index and column index of the matrix, respectively, and N is the number of rows or columns of the gray-level co-occurrence matrix. The contrast ratio is calculated as follows:

[0071]

[0072] Where C represents contrast, (ij) 2 P(i,j) represents the squared difference between the row index and column index of the current element, and P(i,j) is the gray-level co-occurrence matrix element value at the corresponding position. The definition of N is consistent with the formula for texture energy value.

[0073] The spectral and texture feature sets are input into a multi-scale segmentation algorithm. This algorithm employs a region merging strategy, with the initial segmentation unit being a single pixel. The merging rule is that the spectral similarity difference between adjacent pixel blocks is less than a first threshold, and the texture difference is less than a second threshold. The first threshold is dynamically adjusted based on the standard deviation of the Normalized Difference Vegetation Index (NDVI) and the Enhanced Vegetation Index (EVI). For example, when the standard deviation is greater than 0.1, the first threshold is set to 0.05. The second threshold is set based on the range of texture energy values. For example, when the range is greater than 50, the second threshold is set to 10. By iteratively merging adjacent pixel blocks, vegetation cover patches of different spatial scales are generated, forming a multi-scale vegetation distribution map. Large-scale patches are used for watershed-level vegetation control planning, with a merging threshold of spectral similarity difference less than 0.1 and texture difference less than 15. Small-scale patches are used for local vegetation type identification, with segmentation parameters of spectral similarity difference less than 0.02 and texture difference less than 5.

[0074] S2. By analyzing the spatial structure characteristics of multi-scale vegetation distribution maps, and combining topographic and hydrological factors, a spatial correlation network of vegetation measures is constructed, including:

[0075] Based on multi-scale vegetation distribution maps, the aggregation index and proximity index of vegetation cover patches are extracted to generate a spatial structure feature set.

[0076] Obtain digital elevation model data of the study area, calculate slope, terrain curvature and elevation variation coefficient, and generate a set of terrain factors;

[0077] Data on cumulative runoff and river network density in the study area were obtained, and the runoff path of the watershed was simulated based on the hydrological model to generate a set of hydrological factors.

[0078] By inputting the spatial structure feature set, topographic factor set, and hydrological factor set into the spatial network model, and by coupling the spatial adjacency relationship of vegetation patches with the topographic-hydrological interaction weight, a spatial association network of vegetation measures is constructed.

[0079] Based on multi-scale vegetation distribution maps, the aggregation index and proximity index of vegetation cover patches were extracted. The aggregation index was calculated as the ratio of the actual aggregation degree of vegetation patches in the study area to the aggregation degree under random distribution conditions. The proximity index was calculated as the reciprocal of the average distance between adjacent vegetation patches. Specifically, a moving window analysis method was used, with a window size of 1 km × 1 km and a moving step size of 500 m, covering the entire study area to generate a spatial structure feature set. The spatial structure feature set includes an aggregation index matrix and a proximity index matrix for each vegetation patch. The aggregation index ranges from 0 to 1, with values ​​closer to 1 indicating higher aggregation. The proximity index ranges from 0 to positive infinity, with larger values ​​indicating stronger spatial proximity between patches.

[0080] Digital elevation model (DEM) data for the study area was acquired, with the spatial resolution of the DEM data consistent with that of the multi-scale vegetation distribution map. Based on the DEM data, slope, topographic curvature, and elevation coefficient of variation were calculated. Slope was calculated using the third-order inverse distance-squared difference method, calculating the elevation change rate of each grid cell in the x and y directions, and obtaining the slope value by the square root of the sum of squares. Topographic curvature calculation included profile curvature and planar curvature; profile curvature reflected the topographic undulation along the slope direction, while planar curvature reflected the degree of topographic curvature perpendicular to the slope direction. The elevation coefficient of variation was calculated as the ratio of the standard deviation of elevation values ​​within the study area to the average elevation, used to quantify the degree of topographic undulation. A topographic factor set was generated, including a slope matrix, a topographic curvature matrix, and an elevation coefficient of variation matrix, where slope was in degrees, topographic curvature was in meters, and the elevation coefficient of variation was a dimensionless parameter.

[0081] The slope is calculated using the third-order inverse distance squared difference method, and the specific formula is as follows:

[0082]

[0083] Where S is the slope value; The rate of change of elevation of a grid cell in the x-direction is expressed by the following formula:

[0084]

[0085] The rate of change of elevation of a grid cell in the y-direction is expressed by the following formula:

[0086]

[0087] In the formula, Z a,b This represents the elevation value of the raster cell in row a and column b, where a is the row index, b is the column index, Δx and Δy are the spatial resolutions of the raster cell in the x and y directions, respectively, and arctan is the arctangent function. The coefficient for converting radians to degrees.

[0088] The method for calculating the coefficient of variation of elevation is as follows:

[0089]

[0090] Among them, CV H σ is the coefficient of variation of elevation. H The standard deviation of the elevation values ​​of all grid cells within the study area is expressed by the following formula:

[0091]

[0092] μ H The average elevation value of all grid cells within the study area is represented by the following formula:

[0093]

[0094] In the formula, H k This represents the elevation value of the k-th raster cell, where k is the raster cell number, K is the total number of raster cells in the study area, and CV. H is the dimensionless coefficient of variation of elevation.

[0095] Accumulated runoff and river network density data for the study area were acquired. Accumulated runoff was calculated based on a hydrological model simulating the watershed runoff path. The hydrological model used the D8 single-direction algorithm to determine the flow direction. The D8 single-direction algorithm selects the direction with the largest elevation difference by comparing the elevation difference between each grid cell and its eight adjacent grid cells. The upstream catchment area of ​​each grid cell was accumulated based on the flow direction to generate accumulated runoff data, with the unit being square meters. River network density data was calculated as the river length per unit area within the study area. The runoff threshold for extraction was dynamically set based on the study area area; for example, when the study area was greater than 100 square kilometers, the runoff threshold was set to 5000 pixels; when the study area was less than or equal to 100 square kilometers, the runoff threshold was set to 2000 pixels. A set of hydrological factors was generated, including an accumulated runoff matrix and a river network density matrix, with the unit of river network density being kilometers per square kilometer.

[0096] The spatial structure feature set, topographic factor set, and hydrological factor set are input into the spatial network model. The construction rules of the spatial network model are as follows: the spatial adjacency relationship between vegetation patches is defined as the contact of patch boundaries or the Euclidean distance between patches is less than 200 meters; the topographic-hydrological interaction weight is calculated by the reciprocal of the product of slope and runoff accumulation plus a linear combination of elevation variation coefficient and river network density.

[0097] Specifically, the formula for calculating the topographic-hydrological interaction weight is as follows:

[0098] Weight value = 1 / (slope × cumulative runoff) + 0.3 × elevation variation coefficient + 0.2 × river network density

[0099] The reciprocal of the product of slope and runoff accumulation is used to characterize the interaction sensitivity between flat and high-runoff areas, while the linear combination of elevation variation coefficient and river network density reflects the synergistic effect of topographic complexity and river distribution. A spatial association network for vegetation measures is constructed by coupling spatial adjacency relationships with topographic-hydrological interaction weights. Nodes in the spatial association network represent vegetation patches, and edge weights represent the intensity of topographic-hydrological interactions between patches. The edge weights range from 0 to positive infinity, with larger values ​​indicating more significant interactions.

[0100] S3. Perform topological analysis on the spatially interconnected network, and calculate the functional evaluation index of vegetation measures by extracting node degree centrality, betweenness centrality, and clustering coefficients, including:

[0101] Based on the spatial association network, the degree centrality of each node is calculated. The degree centrality is the ratio of the number of edges directly connected to the node to the maximum possible number of edges connected to the network.

[0102] Calculate the betweenness centrality of each node, which is the ratio of the number of times the node is on the shortest path among other nodes to the total number of shortest paths in the network;

[0103] Calculate the clustering coefficient for each node. The clustering coefficient is the ratio of the actual number of connecting edges between the node's neighboring nodes to the theoretical maximum number of connecting edges.

[0104] The degree centrality, betweenness centrality, and clustering coefficient are input into the functional assessment model. The calculation formula of the functional assessment model is the product of degree centrality and betweenness centrality plus the interaction term between clustering coefficient and degree centrality, which generates the functional assessment index of vegetation measures.

[0105] The functional evaluation index is normalized, and the normalized value range is from 0 to 1.

[0106] Based on a spatial association network, the degree centrality of each node is calculated. The degree centrality is calculated as the ratio of the number of edges directly connected to the node to the maximum possible number of edges connected to the node, which is the total number of nodes minus one. For example, when there are N nodes in the spatial association network, the maximum possible number of edges connected to each node is N-1. If a node is actually connected to k edges, then its degree centrality is k / (N-1). A degree centrality matrix is ​​generated by traversing all nodes in the spatial association network. Each element in the matrix corresponds to the degree centrality value of a node, ranging from 0 to 1. A value closer to 1 indicates a higher connection density for the node in the network.

[0107] Calculate the betweenness centrality of each node. Betweenness centrality is calculated as the ratio of the number of times a node lies on the shortest path between other node pairs to the total number of shortest paths in the network. The shortest path calculation uses Dijkstra's algorithm, which iteratively updates the distance labels between nodes, prioritizing the node with the smallest current distance for expansion, until all nodes have been traversed. For example, if a node exists in 100 shortest paths and the total number of shortest paths in the network is 500, then the betweenness centrality of that node is 100 / 500 = 0.2. Generate a betweenness centrality matrix. Each element in the matrix corresponds to the betweenness centrality value of a node, ranging from 0 to 1. The closer the value is to 1, the stronger the node's pivotality in the network.

[0108] Calculate the clustering coefficient for each node. The clustering coefficient is calculated as the ratio of the actual number of connecting edges between a node's neighbors to the theoretical maximum possible number of connecting edges. The theoretical maximum possible number of connecting edges is calculated by multiplying the number of neighboring nodes by the number of neighboring nodes minus one, and then dividing by two. For example, if a node has 5 neighboring nodes and there are actually 3 connecting edges between them, then the theoretical maximum number of connecting edges is 5 × 4 / 2 = 10, and the clustering coefficient is 3 / 10 = 0.3. Generate a clustering coefficient matrix. Each element in the clustering coefficient matrix corresponds to the clustering coefficient value of a node. The clustering coefficient value ranges from 0 to 1, with values ​​closer to 1 indicating a more significant local clustering of the node.

[0109] Input the degree centrality matrix, betweenness centrality matrix, and clustering coefficient matrix into the functional assessment model. The functional assessment model is calculated by multiplying the degree centrality and betweenness centrality, plus the interaction term between the clustering coefficient and degree centrality.

[0110] Specifically, the formula for calculating the functional assessment index is as follows:

[0111] Functional evaluation index = Degree centrality × Betweenness centrality + 0.5 × Clustering coefficient × Degree centrality

[0112] Here, 0.5 is the weighting coefficient for the interaction terms. The weighting coefficient is set based on the historical data analysis results or experimental calibration results in the expert knowledge base, and is used to balance the moderating effect of the clustering coefficient on degree centrality. A functional evaluation index matrix is ​​generated. Each element in the functional evaluation index matrix corresponds to the functional evaluation index value of a node. The functional evaluation index value ranges from 0 to positive infinity. The larger the value, the better the soil and water conservation function of the vegetation measures corresponding to that node.

[0113] The functional assessment index matrix is ​​normalized. The normalization process uses a minimum-maximum method, linearly mapping the functional assessment indices to the interval between 0 and 1. Specifically, the normalized functional assessment index = (original index value - minimum value) / (maximum value - minimum value), where the minimum value is the minimum value in the functional assessment index matrix, and the maximum value is the maximum value in the functional assessment index matrix. The normalized value range is 0 to 1, with values ​​closer to 1 indicating better soil and water conservation function of the vegetation measures. The result of the normalization process is stored as a normalized functional assessment index matrix, which is used for threshold comparison and optimization area identification in subsequent steps.

[0114] S4. Compare the functional assessment index with the preset benchmark threshold to determine whether the vegetation measures meet the soil and water conservation requirements, including:

[0115] Obtain preset benchmark thresholds, which include dynamic quantile thresholds based on historical soil and water conservation data and empirical thresholds from an expert knowledge base;

[0116] The normalized functional assessment index is compared with the dynamic quantile threshold. If the normalized functional assessment index is lower than the 25th percentile of the dynamic quantile threshold, it is determined that the soil and water conservation requirements are not met.

[0117] The normalized functional assessment index is compared with the empirical threshold. If the index value is lower than the lower limit of the empirical threshold and the index value in a local area has decreased continuously for more than three cycles, it is determined that the soil and water conservation requirements are not met.

[0118] Based on the global and local judgment results, a state label for meeting the soil and water conservation requirements is generated. The state label includes three categories: met, critical, and not met.

[0119] Preset baseline thresholds are obtained, including dynamic quantile thresholds based on historical soil and water conservation data and empirical thresholds from an expert knowledge base. Historical soil and water conservation data comes from monitoring data on vegetation cover, soil erosion modulus, and runoff coefficient in the study area over the past ten years. The dynamic quantile thresholds are calculated by dividing the dataset by year and extracting the 25th percentile value of the vegetation measure function assessment index for each year, forming a dynamic quantile threshold sequence. The empirical thresholds in the expert knowledge base are constructed through expert questionnaires and a historical success case database. The lower limit is the lowest acceptable function assessment index agreed upon by experts, for example, a lower limit of 0.4. Both dynamic and empirical thresholds are stored according to the type of study area; for example, the dynamic quantile threshold for mountainous watersheds is 0.35, and for plain watersheds it is 0.25. The dynamic quantile thresholds are updated annually, with the threshold sequence being adjusted based on the latest monitoring data.

[0120] The Normalized Functional Index (NFI) is compared with the dynamic quantile threshold. The matching rule for the dynamic quantile threshold is to select the quantile value corresponding to the study area type and the current year. If the NFI is lower than the 25th percentile of the dynamic quantile threshold, it is determined that the soil and water conservation requirements are not met. For example, if the study area is a mountainous watershed and the dynamic quantile threshold for the current year is 0.35, and the NFI for a certain area is 0.3, then that area is determined to not meet the soil and water conservation requirements. The 25th percentile of the dynamic quantile threshold is calculated and updated dynamically using historical data to ensure that the threshold is dynamically adjusted according to environmental changes. During the determination process, if the study area type does not match in the historical data, the threshold of adjacent similar watersheds is used by default.

[0121] The Normalized Functional Index (NFI) is compared with an empirical threshold. The lower limit of the empirical threshold is set according to the type of study area; for example, the lower limit is 0.4 for mountainous watersheds and 0.3 for plain watersheds. If the NFI is lower than the lower limit of the empirical threshold and the index value in a local area declines for more than three consecutive periods, it is determined that the soil and water conservation requirements are not met. The number of consecutive decline periods is determined through time series analysis. For example, using quarters as the period unit, when the NFI value of a certain area is lower than the value of the previous period for three consecutive quarters, the continuous decline judgment condition is triggered. The time series analysis uses the sliding window method, with a window size of three periods and a step size of one period. The sliding window method is calculated by going back two periods from the current period, and a total of three periods of data are used for analysis. If the index value decreases period by period within the three periods, it is determined to be a continuous decline.

[0122] Status labels for water and soil conservation requirements are generated based on global and local assessment results. The global assessment result is a comparison between the Normalized Functional Assessment Index (NFIA) and the dynamic quantile threshold, while the local assessment result is a comparison between the NFIA and an empirical threshold. The generation rules for status labels are as follows: if both global and local assessments are met, the status label is "Met"; if only one assessment is met, the status label is "Critical"; if neither assessment is met, the status label is "Not Met". Status labels are stored in grid cells, with each grid cell corresponding to one status label value. Status label values ​​include three categories: "Met", "Critical", and "Not Met". The storage format for status labels is a categorized encoding, for example, "Met" is encoded as 1, "Critical" as 2, and "Not Met" as 3. The update frequency of status labels is synchronized with the dynamic quantile threshold, updated annually to ensure consistency between the assessment results and the latest environmental data.

[0123] S5. When the soil and water conservation requirements are not met, construct a causal gradient field for the potential energy field of soil and water conservation functions. Extract the abrupt change points of nodes in the gradient flow through neural information bottleneck compression, and define the priority optimization region, including:

[0124] A potential energy field for soil and water conservation functions is constructed. This potential energy field is generated by the product of the normalized functional assessment index and the topographic-hydrological interaction weight.

[0125] Based on the potential energy field of soil and water conservation function, the causal gradient direction and intensity of each node in the potential energy field are calculated by the gradient descent algorithm to form a causal gradient field.

[0126] The causal gradient field is input into the neural information bottleneck compression model. The neural information bottleneck compression model extracts the mutation points in the gradient flow through a multilayer perceptron. The criteria for determining the mutation point are that the gradient direction mutation angle exceeds the preset angle threshold and the gradient intensity change rate is significantly higher than the historical average level.

[0127] Based on the spatial distribution density and intensity threshold of mutation points, a priority optimization region is defined. The priority optimization region is the region where the mutation point density is higher than the preset density threshold and the average intensity exceeds the preset intensity threshold.

[0128] A potential energy field for soil and water conservation is constructed, which is generated by multiplying the normalized functional assessment index and the topographic-hydrological interaction weight. The normalized functional assessment index is derived from the state label determination result generated in step S4, and the topographic-hydrological interaction weight is derived from the edge weights in the spatial association network constructed in step S2. The physical meaning of the potential energy field is to quantify the potential impact of vegetation measures on soil and water conservation under specific topographic and hydrological conditions. The value range of the potential energy field is limited to 0 to 1 through normalization; the closer the value is to 1, the higher the potential for soil and water conservation. The spatial resolution of the potential energy field is consistent with the resolution of the multi-scale vegetation distribution map. Each grid cell corresponds to a potential energy field value, which is calculated by multiplying the normalized functional assessment index of the corresponding grid cell by the topographic-hydrological interaction weight.

[0129] Based on the potential energy field for soil and water conservation, the causal gradient direction and intensity of each node in the potential energy field are calculated using the gradient descent algorithm. The iteration step size of the gradient descent algorithm is dynamically adjusted according to the local rate of change of the potential energy field value. Specifically, when the local rate of change is greater than a preset threshold, the step size is reduced to 50% of the original step size; when the local rate of change is less than or equal to the preset threshold, the step size remains at 100% of the original step size. The causal gradient direction is calculated as the direction of maximum descent of the potential energy field value in space, and the causal gradient intensity is calculated as the rate of change of the potential energy field value along the gradient direction. A causal gradient field is generated, which includes a gradient direction matrix and a gradient intensity matrix for each node. The gradient direction is represented by an angle value, and the gradient intensity is represented by a dimensionless numerical value. Each element of the gradient direction matrix corresponds to the angle value of a node, and each element of the gradient intensity matrix corresponds to the intensity value of a node.

[0130] A causal gradient field is input into a neural information bottleneck compression model, which extracts abrupt changes in the gradient flow using a multilayer perceptron. The neural information bottleneck compression model consists of an input layer, a hidden layer, and an output layer. The input layer receives gradient direction and intensity data, the hidden layer extracts higher-order features using a nonlinear activation function, and the output layer generates probability values ​​for abrupt changes. Abrupt changes are determined when the gradient direction abrupt change angle exceeds a preset threshold and the gradient intensity change rate is significantly higher than the historical average. The gradient direction abrupt change angle is calculated as the angle between the gradient directions of adjacent nodes. The preset threshold is determined through expert experience or historical data; for example, by analyzing gradient direction abrupt change characteristics in historical successful cases, the preset threshold is set to 30 degrees. The gradient intensity change rate is calculated as the ratio of the current gradient intensity to the historical average gradient intensity. The historical average is derived from the average gradient intensity over the past five years in the study area. The historical average gradient intensity is calculated as the arithmetic mean of the gradient intensity data for each year.

[0131] Priority optimization regions are defined based on the spatial distribution density and intensity threshold of mutation points. The spatial distribution density of mutation points is calculated as the number of mutation points per square kilometer. The preset density threshold is dynamically set according to the study area type; for example, the density threshold is set to 5 per square kilometer for mountainous watersheds and 3 per square kilometer for plain watersheds. The intensity threshold is the average gradient intensity of the mutation points. The preset intensity threshold is determined using a successful case database from an expert knowledge base; for example, by statistically analyzing the average gradient intensity of historical optimization regions, the preset intensity threshold is set to 0.7. Priority optimization regions are defined as areas that simultaneously satisfy both a mutation point density higher than the preset density threshold and an average intensity exceeding the preset intensity threshold. The boundaries of the priority optimization regions are generated using a spatial clustering algorithm. The clustering algorithm used is DBSCAN density clustering, with a neighborhood radius set to 500 meters and a minimum number of neighborhood points set to 10. Generate a priority optimization region map. Each region in the priority optimization region map corresponds to an optimization priority label. The priority labels are divided into three levels: high, medium, and low. The priority label is divided based on the degree of exceeding the density and intensity thresholds. For example, regions where the density exceeds the threshold by more than 50% and the intensity exceeds the threshold by more than 30% are marked as high priority.

[0132] S6. Based on the spatial location and functional requirements of the priority optimization area, generate vegetation measure adjustment suggestions and update the spatial association network, including:

[0133] Based on the spatial location of the priority optimization area, outlier values ​​of vegetation cover type and topographic-hydrological interaction weights within the priority optimization area are extracted.

[0134] Based on the outlier values ​​of vegetation cover type and topographic-hydrological interaction weight, suggestions for adjusting vegetation measures are generated. These suggestions include vegetation type replacement, planting density adjustment, and supplementation of soil and water conservation projects.

[0135] Based on the adjustment recommendations, the topographic-hydrological interaction weights in the spatial association network are updated. The weight update rule is to reduce the weights of abnormal nodes and enhance the connection strength of optimized nodes.

[0136] The updated spatial association network is overlaid with the multi-scale vegetation distribution map to generate an optimized vegetation measure distribution map.

[0137] Based on the spatial location of the priority optimization area, outliers in vegetation cover type and topographic-hydrological interaction weights within the priority optimization area are extracted. Vegetation cover type is obtained through spatial query of multi-scale vegetation distribution maps. The criteria for determining outliers in topographic-hydrological interaction weights are that the weight value is lower than a preset lower limit or higher than a preset upper limit of the historical average weight value for the same period. The historical average weight value is derived from the topographic-hydrological interaction weight data of the same season over the past five years in the study area. Outlier extraction uses the Z-score standardization method, and nodes with an absolute Z-score greater than a preset standardization threshold are considered outliers. The extraction results include a vegetation cover type matrix and a topographic-hydrological interaction weight outlier matrix. Each element in the outlier matrix corresponds to an outlier status label for a node, which includes two categories: "low-weight outlier" and "high-weight outlier".

[0138] Based on vegetation cover type and topographic-hydrological interaction weight anomalies, vegetation adjustment suggestions are generated. The suggestion rules for vegetation type replacement are as follows: if the vegetation cover type in the priority optimization area is a low-water and soil conservation efficiency type and has a low-weight anomaly, replace it with a high-water and soil conservation efficiency vegetation type. The suggestion rules for planting density adjustment are as follows: if the topographic-hydrological interaction weight anomaly is a high-weight anomaly and the vegetation cover type is a high-density planting type, reduce the planting density to a preset reasonable range. The suggestion rules for supplementing water and soil conservation projects are as follows: if the priority optimization area has both a low-weight anomaly and a high erosion risk, supplement with terraced fields or silt-trapping dams. Adjustment suggestions are generated by raster unit, with each raster unit corresponding to a set of suggestion tags. Suggestion tags include "Replace vegetation type A," "Adjust density to B plants / hectare," and "Supplement terraced field projects," etc. The specific vegetation type and project type are generated by matching based on the ecological feature database of the study area.

[0139] Based on the adjustment recommendations, the topographic-hydrological interaction weights in the spatial association network are updated. The weight update rules are as follows: for nodes marked as low-weight anomalies, their topographic-hydrological interaction weights are reduced by a preset ratio; for optimized nodes (such as areas with replaced vegetation types or supplementary projects), their connection strength is increased by a preset ratio. The increase in connection strength is dynamically adjusted according to the soil and water conservation effectiveness level of the vegetation type; for example, the increase in connection strength for vegetation types with high soil and water conservation effectiveness is greater than that for medium-effective types. The updated topographic-hydrological interaction weight matrix is ​​then reintegrated with other topological parameters of the spatial association network to form the updated spatial association network. The updated network retains the original spatial adjacency relationships between nodes and edges, adjusting only the weight values ​​of anomaly nodes and optimized nodes.

[0140] The updated spatial association network is overlaid with a multi-scale vegetation distribution map to generate an optimized vegetation measure distribution map. The overlay method employs spatial raster algebra operations, using the updated topographic-hydrological interaction weights as weighting factors to perform weighted fusion of vegetation cover types in the multi-scale vegetation distribution map. The weighted fusion rule is as follows: if the weighting factor of a raster cell is higher than a preset fusion threshold, the original vegetation cover type is retained; if the weighting factor is lower than the preset fusion threshold, it is replaced with the vegetation type from the adjustment suggestions. The preset fusion threshold is determined using historical cases from an expert knowledge base; for example, based on the weight distribution of the best-performing historical optimization cases, the fusion threshold is set to 0.6. The optimized vegetation measure distribution map includes vegetation type distribution, planting density labeling, and engineering addition locations. The legend of the distribution map indicates the implementation priority of the adjustment suggestions, which is comprehensively evaluated based on the weighting factor and the adjustment range. The priority labels are divided into high, medium, and low levels.

[0141] The calculations involved in the embodiments are all dimensionless numerical calculations, and the preset parameters and thresholds in the calculations are set by those skilled in the art according to the actual situation.

[0142] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, in the form of a computer program product.

[0143] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and inventive constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0144] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0145] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or modules may be electrical, mechanical, or other forms.

[0146] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0147] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for intelligent extraction and functional evaluation of vegetation measures in typical watersheds driven by multi-source remote sensing, characterized in that, Includes the following steps: S1. Acquire multi-source remote sensing image data of the study area, extract vegetation cover information based on spectral and texture features, and perform multi-level segmentation of vegetation cover information to generate multi-scale vegetation distribution maps. S2. By analyzing the spatial structure characteristics of multi-scale vegetation distribution maps, a spatial correlation network of vegetation measures is constructed by combining topographic and hydrological factors. S3. Perform topological analysis on the spatial network and calculate the functional evaluation index of vegetation measures by extracting node degree centrality, betweenness centrality and clustering coefficient. S4. Compare the functional evaluation index with the preset benchmark threshold to determine whether the vegetation measures meet the soil and water conservation requirements. S5. When the soil and water conservation requirements are not met, construct the causal gradient field of the soil and water conservation functional potential field, extract the mutation points of nodes in the gradient flow through neural information bottleneck compression, and define the priority optimization area. S6. Based on the spatial location and functional requirements of the priority optimization area, generate suggestions for adjusting vegetation measures and update the spatial association network.

2. The method for intelligent extraction and functional evaluation of vegetation measures in typical watersheds driven by multi-source remote sensing according to claim 1, characterized in that, Multi-source remote sensing image data of the study area were acquired, and vegetation cover information was extracted based on spectral and texture features. The vegetation cover information was then segmented at multiple levels to generate a multi-scale vegetation distribution map, including: Acquire multispectral and high-resolution panchromatic remote sensing images of the study area, perform radiometric and geometric corrections on the multispectral remote sensing images, and generate preprocessed multi-source remote sensing image data. Based on the preprocessed multispectral remote sensing images, the normalized vegetation index and enhanced vegetation index were calculated, and the spectral characteristics of the vegetation-covered area were extracted. Based on the preprocessed high-resolution panchromatic remote sensing image, the texture energy value and contrast of the vegetation-covered area are extracted using the gray-level co-occurrence matrix algorithm to generate a texture feature set. The spectral and texture feature sets are input into a multi-scale segmentation algorithm. By iteratively merging adjacent pixel blocks, vegetation cover patches of different spatial scales are generated, forming a multi-scale vegetation distribution map.

3. The method for intelligent extraction and functional evaluation of vegetation measures in typical watersheds driven by multi-source remote sensing according to claim 1, characterized in that, By analyzing the spatial structure characteristics of multi-scale vegetation distribution maps, and combining topographic and hydrological factors, a spatial correlation network of vegetation measures is constructed, including: Based on multi-scale vegetation distribution maps, the aggregation index and proximity index of vegetation cover patches are extracted to generate a spatial structure feature set. Obtain digital elevation model data of the study area, calculate slope, terrain curvature and elevation variation coefficient, and generate a set of terrain factors; Data on cumulative runoff and river network density in the study area were obtained, and the runoff path of the watershed was simulated based on the hydrological model to generate a set of hydrological factors. By inputting the spatial structure feature set, topographic factor set, and hydrological factor set into the spatial network model, and by coupling the spatial adjacency relationship of vegetation patches with the topographic-hydrological interaction weight, a spatial association network of vegetation measures is constructed.

4. The method for intelligent extraction and functional evaluation of vegetation measures in typical watersheds driven by multi-source remote sensing according to claim 1, characterized in that, Topological analysis was performed on the spatially interconnected network, and functional evaluation indices of vegetation measures were calculated by extracting node degree centrality, betweenness centrality, and clustering coefficients. These indices include: Based on the spatial association network, the degree centrality of each node is calculated. The degree centrality is the ratio of the number of edges directly connected to the node to the maximum possible number of edges connected to the network. Calculate the betweenness centrality of each node, which is the ratio of the number of times the node is on the shortest path among other nodes to the total number of shortest paths in the network; Calculate the clustering coefficient for each node. The clustering coefficient is the ratio of the actual number of connecting edges between the node's neighboring nodes to the theoretical maximum number of connecting edges. By inputting degree centrality, betweenness centrality, and clustering coefficient into the functional assessment model, a functional assessment index for vegetation measures is generated. The functional evaluation index is normalized, and the normalized value range is from 0 to 1.

5. The method for intelligent extraction and functional evaluation of vegetation measures in a typical watershed driven by multi-source remote sensing according to claim 4, characterized in that, The formula for calculating the functional assessment model is the product of degree centrality and betweenness centrality plus the interaction term between the clustering coefficient and degree centrality.

6. The method for intelligent extraction and functional evaluation of vegetation measures in typical watersheds driven by multi-source remote sensing according to claim 1, characterized in that, The functional assessment index is compared with a preset benchmark threshold to determine whether vegetation measures meet soil and water conservation requirements, including: Obtain preset benchmark thresholds, which include dynamic quantile thresholds based on historical soil and water conservation data and empirical thresholds from an expert knowledge base; The normalized functional assessment index is compared with the dynamic quantile threshold. If the normalized functional assessment index is lower than the 25th percentile of the dynamic quantile threshold, it is determined that the soil and water conservation requirements are not met. The normalized functional assessment index is compared with the empirical threshold. If the index value is lower than the lower limit of the empirical threshold and the index value in a local area has decreased continuously for more than three cycles, it is determined that the soil and water conservation requirements are not met. Based on the global and local judgment results, a state label for meeting the soil and water conservation requirements is generated. The state label includes three categories: met, critical, and not met.

7. The method for intelligent extraction and functional evaluation of vegetation measures in typical watersheds driven by multi-source remote sensing according to claim 1, characterized in that, When soil and water conservation requirements are not met, a causal gradient field of the potential energy field for soil and water conservation functions is constructed. The abrupt change points of nodes in the gradient flow are extracted through neural information bottleneck compression to define priority optimization regions, including: A potential energy field for soil and water conservation functions is constructed. This potential energy field is generated by the product of the normalized functional assessment index and the topographic-hydrological interaction weight. Based on the potential energy field of soil and water conservation function, the causal gradient direction and intensity of each node in the potential energy field are calculated by the gradient descent algorithm to form a causal gradient field. The causal gradient field is input into the neural information bottleneck compression model, which extracts the mutation points in the gradient flow through a multilayer perceptron. Based on the spatial distribution density and intensity threshold of mutation points, a priority optimization region is defined. The priority optimization region is the region where the mutation point density is higher than the preset density threshold and the average intensity exceeds the preset intensity threshold.

8. The method for intelligent extraction and functional evaluation of vegetation measures in typical watersheds driven by multi-source remote sensing according to claim 7, characterized in that, The criteria for determining a mutation point are that the gradient direction mutation angle exceeds a preset angle threshold and the gradient intensity change rate is significantly higher than the historical average level.

9. The method for intelligent extraction and functional evaluation of vegetation measures in typical watersheds driven by multi-source remote sensing according to claim 1, characterized in that, Based on the spatial location and functional requirements of priority optimization areas, vegetation measure adjustment suggestions are generated and the spatial association network is updated, including: Based on the spatial location of the priority optimization area, outlier values ​​of vegetation cover type and topographic-hydrological interaction weights within the priority optimization area are extracted. Based on the outlier values ​​of vegetation cover type and topographic-hydrological interaction weight, suggestions for adjusting vegetation measures are generated. These suggestions include vegetation type replacement, planting density adjustment, and supplementation of soil and water conservation projects. Based on the adjustment recommendations, the topographic-hydrological interaction weights in the spatial association network are updated; The updated spatial association network is overlaid with the multi-scale vegetation distribution map to generate an optimized vegetation measure distribution map.

10. The method for intelligent extraction and functional evaluation of vegetation measures in a typical watershed driven by multi-source remote sensing according to claim 9, characterized in that, The weight update rule is to reduce the weight of abnormal nodes and enhance the connection strength of optimized nodes.

Images