A protective forest monitoring and assessment system based on intelligent visual recognition

Abstract

This invention relates to the field of intelligent monitoring technology for shelterbelts, and discloses a shelterbelt monitoring and assessment system based on intelligent visual recognition. The system includes a multi-source information synchronous acquisition module that captures synchronously perceived information from vision, topography, and meteorology. Through multi-dimensional feature extraction operations, it generates a vegetation growth spatial field containing canopy morphology, topology, and canopy closure change history. Based on this, a degradation feature evolution map is constructed, which can present the outline of degraded patches, their spread trajectory, and a heat map of intensity distribution. This map undergoes deep feature analysis through a trained multi-level degradation recognition network to obtain degradation levels and trend vectors, ultimately forming a comprehensive assessment report with spatiotemporal information. This invention achieves high-precision, automated monitoring and assessment of the dynamic process of shelterbelt degradation.

Description

Technical Field

[0001] This invention relates to the field of intelligent monitoring technology for shelterbelts, specifically a shelterbelt monitoring and evaluation system based on intelligent visual recognition. Background Technology

[0002] Currently, monitoring the degradation of shelterbelts mainly relies on the periodic interpretation of satellite remote sensing imagery and manual ground surveys. Satellite remote sensing technology can acquire vegetation indices over a wide area, but its data is limited by revisit cycles, cloud cover, and spatial resolution, making it difficult to continuously capture subtle changes and instantaneous conditions within the shelterbelts. Manual surveys, on the other hand, are time-consuming, labor-intensive, have limited coverage, and their conclusions are highly subjective. The data obtained by these conventional techniques are discrete in time and singular in dimension, resulting in descriptions of the degradation process remaining at a static, macroscopic level, unable to accurately reflect the spatiotemporal dynamics and intrinsic connections of degradation.

[0003] Existing technical solutions have shortcomings. Different sensors typically operate independently, and the data acquisition timestamps are inconsistent. This leads to temporal misalignment when fusing and analyzing multi-dimensional information such as canopy morphology, three-dimensional surface structure, and microclimate environment. This misalignment results in errors in the established growth models, making it difficult to accurately characterize the causal relationship between specific meteorological events such as drought or wind erosion and the immediate response of forest morphology. Most existing assessment methods output snapshots of the degradation range or level at a certain period, lacking continuous tracking and quantitative expression of how the degradation process spreads spatially and evolves in intensity, resulting in a lack of foresight in management decisions.

[0004] A technical solution is needed to address the challenge of time synchronization of multi-source heterogeneous data at the acquisition end, ensuring the spatiotemporal consistency of the growth status analysis baseline. Simultaneously, a method is required that not only identifies the current state of degradation but also analyzes and visualizes the dynamic trajectory and intensity evolution of degradation from time-series data, thereby enabling a leap from current state assessment to process early warning. Summary of the Invention

[0005] The purpose of this invention is to provide a protective forest monitoring and evaluation system based on intelligent visual recognition to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides a protective forest monitoring and assessment system based on intelligent visual recognition, the system comprising:

[0007] The multi-source information synchronous acquisition module is used to acquire multi-source synchronous sensing information for the target shelterbelt. The multi-source synchronous sensing information is captured by a visual sensing network, a terrain scanning device, and a meteorological monitoring unit at the same time coordinate.

[0008] The multi-dimensional feature extraction module is used to perform multi-dimensional feature extraction operations on the multi-source synchronous sensing information to obtain the vegetation growth spatial field, which includes the canopy morphology topology, canopy closure change history and biomass distribution matrix.

[0009] The degradation map construction module is used to construct a degradation feature evolution map based on the vegetation growth spatial field. The degradation feature evolution map includes the outline of the degradation patch, the degradation spread trajectory, and the degradation intensity distribution heat map.

[0010] The intelligent degradation identification module integrates a trained multi-level degradation identification network to perform deep feature analysis on the degradation feature evolution map, and obtain degradation level determination results and degradation trend vectors.

[0011] The comprehensive assessment report generation module is used to integrate the degradation level determination results and the degradation trend vector to form a comprehensive assessment report on the degradation of protective forests that includes spatial location information and time series information.

[0012] Preferably, the multidimensional feature extraction module performs multidimensional feature extraction operations on the multi-source synchronous sensing information to obtain the vegetation growth spatial field, including the following operations:

[0013] The multi-source synchronous sensing information is aligned and gridded in the spatiotemporal dimension to generate a fusion data cube with a unified spatial reference and timestamp.

[0014] The spectral reflectance sequence, point cloud high-order sequence, and microclimate parameter sequence are separated from the fused data cube;

[0015] The spectral reflectance sequence was used to perform vegetation index inversion calculations to generate leaf area index distribution maps, chlorophyll content distribution maps, and water stress index distribution maps.

[0016] The point cloud elevation sequence is reconstructed and segmented in three dimensions to generate a digital elevation model, a canopy height model, and individual tree segmentation boundaries;

[0017] By integrating the leaf area index distribution map, the digital elevation model, the canopy height model, and the microclimate parameter sequence, the canopy morphology topology, the canopy closure change history, and the biomass distribution matrix are constructed through spatial interpolation and weighted calculation.

[0018] Preferably, the step of performing three-dimensional reconstruction and segmentation on the point cloud elevation sequence to generate a digital elevation model, a canopy height model, and individual tree segmentation boundaries includes the following steps:

[0019] The point cloud high-order sequence is subjected to denoising and filtering to remove outliers and noisy data, resulting in clean point cloud data;

[0020] Based on the clean point cloud data, a three-dimensional surface model is constructed using a triangulation algorithm. The three-dimensional surface model represents the continuous surface of the terrain and canopy.

[0021] Ground point cloud is extracted from the three-dimensional surface model, and a digital elevation model is generated using an interpolation method.

[0022] Non-ground point clouds are separated from the three-dimensional surface model, and the elevation values ​​of the corresponding locations in the digital elevation model are subtracted from the elevation values ​​of the non-ground point clouds to obtain normalized canopy height point clouds.

[0023] The canopy height point cloud is clustered and segmented, and the boundaries of individual tree canopies are identified based on the spatial distance and density characteristics of the point cloud to generate individual tree segmentation boundaries;

[0024] Based on the canopy height point cloud and the single tree segmentation boundary, a canopy height model is generated by statistically analyzing the point cloud height distribution within each segmentation region.

[0025] Preferably, the degradation map construction module constructs a degradation feature evolution map based on the vegetation growth spatial field, including the following steps:

[0026] Regions with missing canopy coverage and regions with abnormally decreased canopy height were identified in the canopy morphology topology, and the identified regions were marked as initial degraded plaques;

[0027] The process of canopy density change is analyzed, the canopy density decay rate and decay direction are calculated, and the degradation and spread trajectory is simulated based on the decay rate and decay direction.

[0028] Locate regions with significantly low biomass values ​​in the biomass distribution matrix, calculate the biomass gradient difference between the regions with significantly low biomass values ​​and the surrounding regions, and generate the degradation intensity distribution heatmap based on the biomass gradient difference.

[0029] The spatial coordinates of the initial degraded patch, the simulated path of the degradation spread trajectory, and the intensity values ​​of the degradation intensity distribution heatmap are layered and correlated to form the degradation feature evolution map.

[0030] Preferably, the step of analyzing the canopy density change process, calculating the canopy density decay rate and decay direction, and simulating the degradation and spread trajectory based on the decay rate and decay direction includes the following steps:

[0031] Extract canopy closure raster data from multiple time points from the canopy closure change history, calculate the difference in canopy closure values ​​between adjacent time points, and obtain the canopy closure change raster.

[0032] Trend analysis was performed on the canopy density change grid. The slope of the canopy density change of each grid cell over time was fitted using a linear regression method, and the slope of the change was used as the canopy density decay rate.

[0033] Based on the spatial distribution of canopy closure decay rate, the rate gradient of the surrounding area of ​​each grid cell is calculated, and the direction of the fastest rate decrease is determined as the canopy closure decay direction.

[0034] Centered on the initial degraded plaque, a cellular automata model is used to simulate the expansion path of the degraded plaque based on the canopy closure decay rate and canopy closure decay direction. The expansion path takes into account the continuity of rate magnitude and direction.

[0035] The simulated expansion path is verified and adjusted against historical degraded patches to ultimately generate the degradation spread trajectory.

[0036] Preferably, the intelligent degradation identification module obtains the degradation level determination result and degradation trend vector, including the following steps:

[0037] The multi-level degradation identification network performs shallow feature extraction on the degradation feature evolution map to obtain the shape complexity, edge fragmentation and spatial clustering of the degradation patches;

[0038] The multi-level degradation identification network extracts mid-level features from the degradation propagation trajectory to obtain the trajectory's tortuosity, expansion speed, and directional stability.

[0039] The multi-level degradation identification network performs deep feature extraction on the degradation intensity distribution heatmap to obtain the centrality, heterogeneity, and spatial autocorrelation of intensity changes;

[0040] The shape complexity, the expansion speed, and the centrality of the intensity change are fused and mapped to output the degradation level determination result.

[0041] The spatial clustering, directional stability, and spatial autocorrelation are extrapolated over time to output the degradation trend vector describing the future direction of change.

[0042] Preferably, the system further includes multi-scale environmental compensation for the degradation level determination result, including:

[0043] Historical climate sequence and soil property data of the target shelterbelt area are obtained, and the environmental carrying capacity is calculated.

[0044] At the landscape scale, the overall degradation index in the degradation level determination result is compensated and corrected based on the environmental carrying capacity.

[0045] At the patch scale, the degradation level of individual degraded patches is adjusted differentially based on the soil attribute data, with the degradation level of areas with lower soil fertility being increased and the degradation level of areas with higher soil fertility being decreased.

[0046] At the pixel scale, the local intensity values ​​in the degradation intensity distribution heatmap are normalized for illumination conditions by incorporating the terrain shadow coefficient.

[0047] Output the final degradation level after multi-scale environmental compensation.

[0048] Preferably, at the landscape scale, the overall degradation index in the degradation level determination result is compensated and corrected based on the environmental carrying capacity, specifically as follows:

[0049] The ratio of the environmental carrying capacity to the standard reference capacity is calculated to obtain the environmental adjustment coefficient;

[0050] Multiply the overall degradation index by the environmental adjustment coefficient to obtain the environmentally corrected overall degradation index;

[0051] When the environmental carrying capacity is lower than a preset threshold, the drought stress enhancement module is activated. The drought stress enhancement module further corrects the overall degradation index after environmental correction based on the water deficit.

[0052] Preferably, the comprehensive assessment report generation module integrates the degradation level determination result and the degradation trend vector to form a comprehensive assessment report on the degradation of protective forests that includes spatial location information and time series information, including the following steps:

[0053] The degradation level determination results are spatially correlated with the administrative division layer and small plot layer in the geographic information system to generate a degradation level distribution map with administrative affiliation and forest tenure information;

[0054] Discretize the degradation trend vector on the time axis to generate a set of degradation state prediction maps for multiple future time nodes;

[0055] Extract the boundaries and attribute information of the degradation areas that urgently need intervention from the degradation level distribution map, and extract the time window of degradation acceleration from the degradation state prediction map set;

[0056] The core content of the comprehensive assessment report on the degradation of protective forests is formed by combining the boundaries of the degraded areas that urgently require intervention, the attribute information, and the time window of accelerated degradation.

[0057] Preferably, the system further includes generating an ecological restoration plan based on the comprehensive assessment report of shelterbelt degradation, including:

[0058] Based on the degradation level of different regions in the degradation level distribution map, a preset set of restoration measures is obtained by matching them with the preset restoration measures library.

[0059] Based on the degradation acceleration time window in the degradation state prediction map set, the start time and duration of implementation are planned for each measure in the preliminary recovery measure set;

[0060] Combining the soil property data and the microclimate parameter sequence, the parameters of the measures in the preliminary restoration measure set are optimized for localization. The measures parameters include vegetation configuration density, irrigation quota and fertilizer application amount.

[0061] The output includes an ecological restoration plan with optimized measures, implementation timeline, and parameter configuration.

[0062] Compared with the prior art, the beneficial effects of the present invention are:

[0063] By establishing a unified high-precision time coordinate and forcibly aligning the acquisition times of visual, topographic, and meteorological data, the distortion problem caused by time lag in the fusion analysis of multi-source heterogeneous data is resolved. This scheme ensures that the canopy image at each moment is strictly bound to the corresponding 3D point cloud and microclimate parameters, constructing a spatiotemporally consistent vegetation growth spatial field. The extracted features, such as canopy morphology topology and canopy closure change history, possess true physical synchronicity, enabling precise correlation between environmental disturbances and the immediate response of tree morphology, thus improving the accuracy of subsequent degradation attribution analysis.

[0064] Based on the spatiotemporally synchronized vegetation growth spatial field, an evolutionary atlas is constructed, incorporating degradation spread trajectories and heatmaps of degradation intensity distribution. This scheme connects the degradation states at discrete time points into a continuous dynamic process. The spread trajectory visually reveals the spatial expansion direction, path, and velocity of degraded patches; the intensity distribution heatmap uses gradient colors to represent the spatiotemporal evolution of degradation severity. This representation advances degradation analysis from static level determination to a machine-analyzable description of dynamic processes, enabling the output degradation trend vector to quantitatively characterize key dynamic parameters of the degradation process, supporting forward-looking ecological risk early warning and precise governance planning. Attached Figure Description

[0065] Figure 1 This is a schematic diagram illustrating the working principle of the protective forest monitoring and evaluation system based on intelligent visual recognition described in this invention.

[0066] Figure 2 A flowchart for generating a vegetation growth spatial field for multidimensional feature extraction operations;

[0067] Figure 3 A flowchart simulating the trajectory of degradation and spread;

[0068] Figure 4 A diagram illustrating the characteristic fusion and grading analysis of degraded patches in protective forests;

[0069] Figure 5 A time-series heat map for predicting the degradation levels of different areas of the protective forest. Detailed Implementation

[0070] 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 skilled in the art without creative effort are within the scope of protection of the present invention.

[0071] Please see Figure 1 This invention provides a protective forest monitoring and assessment system based on intelligent visual recognition. The system includes: a multi-source information synchronous acquisition module responsible for coordinating a visual sensing network, terrain scanning device, and meteorological monitoring unit deployed in the target protective forest area, ensuring that these devices are clock-synchronized based on a unified time server, thereby capturing visible light and multispectral images, laser point cloud data, and meteorological information such as temperature, humidity, and wind speed with the same time coordinates. A multi-dimensional feature extraction module receives the aforementioned multi-source synchronous sensing information, first performing data registration and gridding in the spatiotemporal dimension to form a fused data cube, and then extracting spectral, three-dimensional structure, and environmental parameter sequences from it. Through a series of inversion, reconstruction, and fusion calculations, a vegetation growth spatial field containing canopy morphology topology, canopy closure change history, and biomass distribution matrix is ​​finally generated.

[0072] The degradation map construction module takes the vegetation growth spatial field as input. By identifying canopy gaps and abnormal areas, analyzing the spatiotemporal trends of canopy closure, locating low-biomass areas and calculating their gradients, it outlines degraded patches, simulates degradation spread trajectories, and renders a heatmap of degradation intensity distribution. These elements are integrated to construct a comprehensive degradation feature evolution map that reflects the spatial pattern and dynamics of degradation. The intelligent degradation identification module integrates a multi-level degradation identification network pre-trained with a large number of labeled samples. This network performs feature analysis on the input degradation feature evolution map from shallow to deep, extracting multi-level deep features from patch morphology to spread dynamics and intensity distribution. Through fully connected layer classification and regression, it outputs the degradation level judgment result corresponding to the current state and a degradation trend vector indicating the future direction of change. The comprehensive assessment report generation module receives the output of the intelligent degradation identification module, combines the degradation level and trend information with the spatial base data in the geographic information system, generates a degradation level distribution map labeled with spatial location attributes, and performs time extrapolation on the degradation trend vector to generate a prediction atlas. Then, it extracts key conclusions from these spatial and temporal information products and automatically compiles them into a structured comprehensive assessment report on the degradation of protective forests.

[0073] Example 1: See Figure 2 The multi-dimensional feature extraction module performs multi-dimensional feature extraction operations on multi-source synchronous sensing information to obtain the vegetation growth spatial field. Specifically, this includes the following operations: Aligning and gridding the multi-source synchronous sensing information in the spatiotemporal dimension; using georegistration technology to unify images, point clouds, and meteorological data acquired by different sensors to the same spatial coordinate system and grid cells of the same size; and assigning a consistent timestamp to each grid cell, thereby generating a fused data cube with a unified spatial reference and timestamp. Separating the spectral reflectance sequence, point cloud elevation sequence, and microclimate parameter sequence from the fused data cube based on data attributes. Performing vegetation index inversion calculations on the spectral reflectance sequence; calculating the normalized vegetation index, photochemical vegetation index, etc., based on a preset spectral band combination formula; and further generating leaf area index distribution maps, chlorophyll content distribution maps, and water stress index distribution maps through empirical or physical model inversion. Performing three-dimensional reconstruction and segmentation on the point cloud elevation sequence to generate a digital elevation model, canopy height model, and individual tree segmentation boundaries. By integrating leaf area index distribution maps, digital elevation models, canopy height models, and microclimate parameter sequences, data from different sources were unified to the same grid scale using Kriging spatial interpolation. Based on the weights of each factor's influence on vegetation growth, a weighted superposition calculation was performed to construct a canopy morphology topology representing three-dimensional morphological structure, a canopy closure change history representing temporal changes in coverage, and a biomass distribution matrix representing the space of organic matter accumulation.

[0074] The process involves three-dimensional reconstruction and segmentation of the point cloud elevation sequence to generate a digital elevation model (DEM), a canopy height model, and individual tree segmentation boundaries. The steps include: 1. Denoising and filtering the point cloud elevation sequence using a statistical outlier removal algorithm and voxel-based downsampling filtering to remove outliers and noise, resulting in clean point cloud data. 2. Constructing a three-dimensional surface model based on the clean point cloud data using a triangular meshing algorithm. This surface model represents the continuous surface of the terrain and canopy. 3. Extracting ground point clouds from the three-dimensional surface model using a cloth simulation filtering method, and then interpolating the ground point clouds using an inverse distance weighted interpolation method to generate a DEM. 4. Separating non-ground point clouds from the three-dimensional surface model, and subtracting the corresponding elevation value from the DEM for the height of each point in the non-ground point cloud to obtain a normalized canopy height point cloud. 5. Performing Euclidean clustering segmentation on the canopy height point cloud, setting thresholds based on the spatial distance and density characteristics of the point cloud, identifying and segmenting point cloud clusters of individual tree canopies, and generating individual tree segmentation boundaries. Based on the canopy height point cloud and the segmentation boundary of individual trees, the canopy height model is generated by statistically analyzing the point cloud height distribution within each segmentation region and calculating its percentile height or average height.

[0075] In practical implementation, the multidimensional feature extraction module receives synchronous sensing information from the multi-source information synchronous acquisition module. The multidimensional feature extraction module aligns and grids the synchronous sensing information in the spatiotemporal dimension, uses known control points and digital elevation models to perform geometric correction on visible light images and multispectral images, matches the point cloud elevation sequence acquired by lidar with the image through geographic coordinates, and interpolates the microclimate parameter sequence recorded by the meteorological monitoring unit onto the same spatial grid. All data are unified into grid cells with a resolution of 0.5 meters, and each grid cell is assigned a precise timestamp obtained from a unified time server, generating a fused data cube with a unified spatial reference and timestamp. The fused data cube is a multidimensional array whose dimensions include longitude, latitude, time, spectral band, elevation, and meteorological parameters.

[0076] In some embodiments, a spectral reflectance sequence, a point cloud elevation sequence, and a microclimate parameter sequence are separated from the fused data cube based on data attributes. The spectral reflectance sequence is a set of reflectance values ​​of multiple bands extracted from the spectral dimension of the fused data cube and arranged in time. The point cloud elevation sequence is a time series of a set of data points containing three-dimensional spatial coordinates and reflectance intensity information extracted from the elevation dimension of the fused data cube. The microclimate parameter sequence is a time series of parameters such as air temperature, air humidity, and soil temperature extracted from the meteorological dimension of the fused data cube and located on grid cells. In practice, vegetation index inversion calculations are performed on the spectral reflectance sequence. Based on the reflectance values ​​of the red and near-infrared bands, the normalized vegetation index (NVI) for each grid cell is calculated. The formula for calculating the NVI is defined as (near-infrared reflectance - red reflectance) divided by (near-infrared reflectance + red reflectance). Using a pre-established empirical conversion model between leaf area index and NVI, the NVI distribution map is converted into a leaf area index distribution map. The chlorophyll content distribution map is generated by inverting the reflectance ratio of the red and green bands through a lookup table. The normalized water index is calculated using the reflectance of the near-infrared and short-wave infrared bands to generate a water stress index distribution map.

[0077] Optionally, the point cloud elevation sequence is subjected to 3D reconstruction and segmentation to generate a digital elevation model, a canopy height model, and individual tree segmentation boundaries. The point cloud elevation sequence is then denoised and filtered. Outliers with excessively large distances from their neighborhood averages are identified and removed using a statistical standard deviation multiple method. A voxel mesh downsampling filter is used to reduce point cloud density while preserving its shape, resulting in clean point cloud data. Based on this clean point cloud data, a 3D surface model is constructed using the Delaunay triangulation algorithm. This 3D surface model represents the continuous surface of the terrain and canopy. Ground point clouds are extracted from the 3D surface model using a progressive triangulation filtering method. Kriging interpolation is then used to refine the extracted ground point cloud data. Spatial interpolation is performed on the point cloud to generate a digital elevation model (DEM). Non-ground point clouds are separated from the 3D surface model by setting a height threshold. The original height value of each point in the non-ground point cloud is subtracted from the elevation value of the DEM at the corresponding plane coordinate position to obtain a normalized canopy height point cloud. The canopy height point cloud is then clustered and segmented based on Euclidean distance. A distance threshold is set according to the 3D spatial distance between point clouds and the local point cloud density characteristics to identify the boundaries of individual tree canopies and generate individual tree segmentation boundaries. Based on the canopy height point cloud and individual tree segmentation boundaries, the height distribution of the point cloud in each segmentation region is statistically analyzed to calculate the 95th percentile height value, thus generating a canopy height model.

[0078] It is understandable that by integrating leaf area index (LAI) distribution maps, digital elevation models (DEMs), canopy height models, and microclimate parameter sequences, and through spatial interpolation and weighted calculations, a canopy morphology topology, canopy closure change history, and biomass distribution matrix are constructed. The LAI distribution maps and canopy height models are spatially resampled to ensure they have identical grid cell sizes and spatial ranges. The normalized canopy height value of the same grid cell is multiplied by the LAI value to generate a composite index layer, which characterizes the three-dimensional morphological structure of the canopy. This composite index layer is defined as the canopy morphology topology. The maximum LAI value for each grid cell over time is calculated, and the date of the maximum value is taken as the date when the canopy closure of that cell reaches its peak. The peak date distribution of all grid cells in the entire region is serialized to form a canopy closure change history. Using the LAI distribution maps and canopy height models, combined with air temperature and soil moisture from the microclimate parameter sequence, these are input into the biomass estimation model. The model outputs the estimated biomass value for each grid cell. The estimated biomass values ​​of all grid cells are arranged into a matrix to generate a biomass distribution matrix.

[0079] Example 2: See Figure 3 The degradation map construction module constructs a degradation feature evolution map based on the vegetation growth spatial field, including the following steps: In the canopy morphology topology, by setting height and density thresholds, areas with missing canopy cover and areas with abnormally decreasing canopy height are identified, and the identified continuous areas are marked as initial degradation patches. The canopy closure change history is analyzed, the canopy closure decay rate and decay direction are calculated, and the degradation spread trajectory is simulated based on the decay rate and decay direction. In the biomass distribution matrix, areas with significantly low biomass values ​​are located by percentile comparison, and the average gradient difference of biomass between areas with significantly low biomass values ​​and surrounding buffer areas is calculated. Based on the magnitude of the biomass gradient difference, a degradation intensity distribution heatmap is generated using color gradients. The spatial coordinates of the initial degradation patches, the simulated path of the degradation spread trajectory, and the intensity values ​​of the degradation intensity distribution heatmap are layered and spatially correlated to form a degradation feature evolution map integrating degradation location, dynamics, and intensity within the same spatial framework.

[0080] Canopy closure raster data at multiple equally spaced time points were extracted from the canopy closure change history. The difference in canopy closure values ​​at adjacent time points was calculated to obtain the canopy closure change raster. Trend analysis was performed on the canopy closure change raster, and a linear regression method was used to fit the slope of the canopy closure change over time for each raster cell. The slope of the change was taken as the canopy closure decay rate. Based on the spatial distribution of the canopy closure decay rate, the rate gradient within the neighborhood window surrounding each raster cell was calculated, and the direction of the fastest rate decrease was determined as the canopy closure decay direction. Centered on the initial degraded patch, a cellular automata model was used to simulate the expansion path of the degraded patch according to the magnitude and direction of the canopy closure decay rate. This expansion path considered the continuity of rate magnitude and direction. The simulated expansion path was superimposed with historical degraded patches for verification. The path was smoothed and the connections were adjusted to finally generate the degradation spread trajectory.

[0081] In practical implementation, the degradation map construction module constructs a degradation feature evolution map based on the vegetation growth spatial field. The degradation map construction module reads the vegetation growth spatial field generated by the multi-dimensional feature extraction module, which includes canopy morphology topology, canopy closure change history, and biomass distribution matrix. In the canopy morphology topology, abnormal areas are identified by setting canopy coverage threshold and height reduction threshold. Continuous grid cell areas with canopy coverage values ​​lower than the preset coverage threshold are marked as canopy coverage missing areas. Continuous grid cell areas with the difference between the current period and the historical baseline canopy height model exceeding the preset height reduction threshold are marked as canopy height abnormal reduction areas. The identified canopy coverage missing areas and canopy height abnormal reduction areas are spatially merged and marked as initial degradation patches.

[0082] Optionally, centered on the initial degraded patch, a cellular automata model is used to simulate the expansion path of the degraded patch based on the magnitude and direction of the canopy closure decay rate. The cellular automata model discretizes the space into grid cells consistent with the canopy closure raster data. Each cell is in the state of "healthy", "degraded", or "potentially degraded". The state transition rule of the cell in the next time step is jointly determined by the current canopy closure decay rate of the cell, the canopy closure decay direction of the adjacent cells, and the current state of the cell. The simulation of the expansion path considers the continuity of the rate magnitude and direction. The simulated expansion path is superimposed and verified with the boundaries of historical degraded patches recorded in historical images. The path is adjusted by morphological dilation, erosion, and skeleton extraction operations, and finally a degradation spread trajectory reflecting the possible development path of degradation is generated.

[0083] It can be understood that the process involves locating areas with significantly low biomass values ​​in the biomass distribution matrix, calculating the biomass gradient difference between these areas and their surrounding regions, and generating a heatmap of degradation intensity distribution based on this gradient difference. The value of each grid cell in the biomass distribution matrix is ​​compared with the median or average biomass value of the entire region. Continuous regions with values ​​below a certain percentage of the median or average are identified as areas with significantly low biomass values. A fixed-distance annular buffer zone is created outward from the boundary of these areas. The difference between the average biomass value within these areas and the average biomass value within the annular buffer zone is calculated to obtain the biomass gradient difference. Based on the magnitude of this gradient difference, a continuous color band from green to red is used for mapping; regions with larger gradient differences are more reddish, thus generating a heatmap of degradation intensity distribution that visually represents the differences in degradation intensity. In practice, the spatial coordinates of the initial degraded patches, the simulated path of the degradation spread trajectory, and the intensity values ​​of the degradation intensity distribution heat map are overlaid and correlated. The three are then loaded into the same map view through a geographic information system to ensure that the spatial reference is completely consistent. A spatial index is established so that clicking on any location can simultaneously view the patch identifier, trajectory direction, and heat intensity at that location, forming a degradation feature evolution map that integrates the spatial location, dynamic trend, and severity of degradation.

[0084] Example 3: The intelligent degradation identification module obtains degradation level determination results and degradation trend vectors, including the following steps: A multi-level degradation identification network performs shallow feature extraction on the degradation feature evolution map, extracting the shape complexity, edge fragmentation, and spatial clustering of degradation patches through convolutional and pooling layer operations. The multi-level degradation identification network performs mid-level feature extraction on the degradation spread trajectory, extracting the tortuosity, expansion speed, and directional stability of the trajectory through graph convolutional operations. The multi-level degradation identification network performs deep feature extraction on the degradation intensity distribution heatmap, extracting the centrality, heterogeneity, and spatial autocorrelation of intensity changes through attention-enhanced convolutional layers. The shape complexity, expansion speed, and centrality of intensity changes are fused and mapped using a fully connected layer to output the degradation level determination result. The spatial clustering, directional stability, and spatial autocorrelation are input into the temporal prediction module for temporal extrapolation calculation, outputting a degradation trend vector describing the future direction of change.

[0085] In practical implementation, the trained multi-level degradation recognition network integrated within the intelligent degradation recognition module performs deep feature analysis on the degradation feature evolution map. The multi-level degradation recognition network receives the degradation feature evolution map from the degradation map construction module as input. The degradation feature evolution map is a comprehensive image containing layer information such as the degraded patch outline, degradation spread trajectory, and degradation intensity distribution heatmap. The multi-level degradation recognition network performs shallow feature extraction on the degradation feature evolution map. Through the first set of convolutional layers and max pooling layers, low-level visual features are extracted from the input image. These features are further processed by the shape descriptor algorithm to calculate the shape complexity, edge fragmentation, and spatial clustering of the degradation patches. The shape complexity is quantified by calculating the ratio of the perimeter to the area of ​​the patch outline and the fractal dimension. The edge fragmentation is evaluated by statistically analyzing the frequency of tortuosity changes of the patch edges per unit area. The spatial clustering is measured by calculating the Moran index of the patch's spatial distribution.

[0086] In some embodiments, the multi-level degradation recognition network performs mid-level feature extraction on the degradation propagation trajectory. The degradation propagation trajectory is converted into rasterized orientation and velocity field images in the form of vector paths and input into the network's second feature extraction branch. This branch contains graph convolutional layers for processing trajectory data with topological connections. The graph convolutional layers operate on the neighborhood of trajectory nodes, aggregating node features to learn the global and local patterns of the trajectory. The tortuosity, propagation velocity, and directional stability of the trajectory are parsed from the output of this branch. The tortuosity is obtained by calculating the ratio of the total length of the trajectory path to the straight-line distance between the start and end points. The propagation velocity is calculated by the ratio of the spatial displacement of the trajectory at different time steps to the time interval. The directional stability is measured by statistically analyzing the standard deviation of the trajectory tangent direction angle. In practical implementation, the multi-level degradation identification network performs deep feature extraction on the degradation intensity distribution heatmap. The degradation intensity distribution heatmap is input into the third feature extraction branch of the network in the form of a single-channel intensity value matrix. This branch contains a deep network composed of multiple residual blocks and a self-attention mechanism. The self-attention mechanism assigns different computational weights to different regions of the intensity distribution heatmap to capture long-range spatial dependencies. From the high-level feature map output by the deep network, the centrality, heterogeneity, and spatial autocorrelation of intensity changes are calculated. Centrality is determined by identifying the high-value core region of the intensity distribution and calculating its offset distance from the geometric center. Heterogeneity is measured by calculating the statistical variance of the intensity values ​​within a local window. Spatial autocorrelation is evaluated by calculating the global Moran index to assess the spatial clustering pattern of the entire heatmap.

[0087] Optionally, the shape complexity, expansion rate, and centrality of intensity change are fused and mapped to output a degradation level determination result. The shape complexity feature vector, expansion rate feature vector, and centrality feature vector from the shallow, medium, and deep feature extraction branches are concatenated into a comprehensive feature vector. This comprehensive feature vector is input into a fully connected layer classifier, which maps the comprehensive feature vector to a discrete degradation level category. The degradation level categories include "no degradation," "mild degradation," "moderate degradation," and "severe degradation," and their mapping function is... for:

[0088]

[0089] in: This represents the combined feature vector after concatenation. and These represent the weight matrix and bias vector of the fully connected layer classifier, respectively. The operation selects the category with the highest output probability as the result of the degradation level determination.

[0090] It can be understood that spatial clustering, directional stability, and spatial autocorrelation are extrapolated over time to output a degradation trend vector describing the future direction of change. Scalars of spatial clustering, directional stability, and spatial autocorrelation from different levels are combined into a trend feature vector. This trend feature vector is input into a regression prediction module composed of long short-term memory (LSTM) network units. Based on the change patterns of these features in historical time-series data, the LSM network units predict the feature values ​​for a predetermined time period in the future. The output of the prediction is a multi-dimensional vector, defined as the degradation trend vector. Each dimension of the degradation trend vector corresponds to the predicted change in features such as spatial clustering, directional stability, and spatial autocorrelation at different future time points.

[0091] See Figure 4This is a feature fusion and grading analysis chart of degraded patches in shelterbelts, showcasing the multidimensional features of different degraded patches and the correlation between their comprehensive scores and degradation levels. As the comprehensive feature score increases with patch number (P01→P04), the corresponding shape complexity, expansion rate, and centrality also increase, with the degradation level upgrading from "non-degraded" (P01) to severely degraded (P04). P05 is a "non-degraded" patch, with its comprehensive score and various feature values ​​at relatively low levels. From P06 to P08, the comprehensive score and feature values ​​increase again, with the degradation level gradually rising. The comprehensive feature score is the result of multidimensional feature fusion; a higher score corresponds to a more severe degradation level, reflecting that "complex shape, rapid expansion, and high centrality" are the core characteristics of severe degradation. This chart is used for intelligent identification and grading of degraded patches in shelterbelts: by correlating multidimensional features with the comprehensive score, severely degraded patches can be quickly located, providing data for subsequent degradation trend analysis and ecological restoration plan development.

[0092] Example 4: This method also includes multi-scale environmental compensation for the degradation level determination results. Historical climate sequences and soil property data of the target shelterbelt area are acquired, and the environmental carrying capacity is calculated using a pre-set environmental carrying capacity model. At the landscape scale, the overall degradation index in the degradation level determination results is compensated and corrected based on the environmental carrying capacity. At the patch scale, the degradation level of individual degraded patches is differentiated based on soil property data, with the degradation level of areas with lower soil fertility being increased and the degradation level of areas with higher soil fertility being decreased. At the pixel scale, the local intensity values ​​in the degradation intensity distribution heatmap are normalized for illumination conditions using the topographic shading coefficient. The final degradation level after multi-scale environmental compensation is output. At the landscape scale, the overall degradation index in the degradation level determination results is compensated and corrected based on the environmental carrying capacity, specifically by calculating the ratio of the environmental carrying capacity to the standard reference capacity to obtain the environmental adjustment coefficient. The overall degradation index is multiplied by the environmental adjustment coefficient to obtain the environmentally corrected overall degradation index. When the environmental carrying capacity is lower than the preset threshold, the drought stress enhancement module is activated. The drought stress enhancement module further corrects the overall degradation index after environmental correction based on the water deficit.

[0093] In its implementation, the system employs a multi-scale environmental compensation method based on the degradation level determination results from the intelligent degradation identification module. It acquires historical climate sequences and soil property data for the target shelterbelt area. The historical climate sequences are derived from long-term observation records of regional meteorological stations, including monthly average precipitation and monthly average temperature sequences for more than ten years. The soil property data are obtained from field sampling and laboratory analysis reports, including information on soil organic matter content, soil field water holding capacity, and soil bulk density. The environmental carrying capacity is calculated using a pre-set environmental carrying capacity model. The environmental carrying capacity model takes the hydrothermal conditions of the historical climate sequence and the fertility and water holding capacity of the soil property data as input parameters, and calculates a dimensionless index, i.e., environmental carrying capacity, through weighted comprehensive calculation.

[0094] In some embodiments, at the landscape scale, the overall degradation index in the degradation level determination result is compensated and corrected based on the environmental carrying capacity. The overall degradation index is a general score of the degradation status of the entire target shelterbelt area output by the intelligent degradation identification module. The ratio of the environmental carrying capacity to a standard reference capacity is calculated. The standard reference capacity is taken from the calculated environmental carrying capacity of the undisturbed top vegetation community in the same climate zone. The obtained ratio is defined as the environmental adjustment coefficient. The overall degradation index is multiplied by the environmental adjustment coefficient to obtain the environmentally corrected overall degradation index. The calculation formula is as follows:

[0095]

[0096] in: This represents the overall degradation index after environmental correction. Represents the original overall degradation index. This represents the environmental regulation coefficient. When the environmental carrying capacity is lower than a preset threshold, the drought stress enhancement module is activated. The preset threshold is set according to a certain percentage of the region's long-term average precipitation. The drought stress enhancement module further corrects the overall degradation index after environmental correction based on the water deficit. The water deficit is calculated by comparing the actual evapotranspiration with the potential evapotranspiration. The correction process uses a linear or nonlinear enhancement function, so that under the same vegetation degradation performance, the final degradation level assessment of areas with more severe drought stress is higher.

[0097] Optionally, at the patch scale, the degradation level of individual degraded patches is differentially adjusted based on soil attribute data. Soil attribute data is spatially interpolated to form a soil fertility distribution map registered with the degraded patch layer. The average soil organic matter content at each degraded patch location in the soil fertility distribution map is compared with the average value for the entire region. The degradation level of areas with lower soil fertility is increased by one level from the original assessment. At the pixel scale, the local intensity values ​​in the degradation intensity distribution heatmap are normalized for illumination conditions using the topographic shading coefficient. The topographic shading coefficient is generated using a digital elevation model and solar azimuth and elevation angles, reflecting the differences in sunlight caused by topography. The original intensity value of each pixel in the degradation intensity distribution heatmap is divided by the corresponding topographic shading coefficient to eliminate some of the influence of topographic shading on the canopy's apparent degradation. The final degradation level after multi-scale environmental compensation is output. It can be understood that the multi-scale environmental compensation process involves the comprehensive application of data at different spatial scales. Refer to Table 1 to show the key parameter relationships in landscape-scale compensation.

[0098] Table 1: Schematic Table of Environmental Compensation Parameters at the Landscape Scale

[0099] Area code Original Overall Degradation Index Environmental carrying capacity Standard reference capacity Environmental regulation coefficient Overall Degradation Index after Environmental Correction Area A 65 0.85 1.20 0.71 46.2 Area B 65 1.10 1.20 0.92 59.8 Area C 65 0.50 1.20 0.42 27.3

[0100] In practice, after multi-scale environmental compensation, the system outputs a final degradation level distribution data that integrates overall landscape-scale correction, patch-scale differential adjustment, and pixel-scale illumination normalization processing results. This data serves as the basis for generating subsequent assessment reports and restoration plans.

[0101] Example 5: The comprehensive assessment report generation module integrates degradation level determination results and degradation trend vectors to form a comprehensive assessment report on shelterbelt degradation that includes spatial location information and time series information. This includes the following steps: Spatially correlate and overlay the degradation level determination results with the administrative division layer and sub-plot layer in the geographic information system to generate a degradation level distribution map with administrative affiliation and forest tenure information. Discretize the degradation trend vector on the time axis and iteratively predict it at set time steps to generate a degradation state prediction map set for multiple future time nodes. Extract the boundaries and attribute information of degradation areas requiring intervention from the degradation level distribution map, and extract the time windows of accelerated degradation from the degradation state prediction map set. Combine the boundaries and attribute information of degradation areas requiring intervention and the time windows of accelerated degradation to form the core content of the comprehensive assessment report on shelterbelt degradation.

[0102] This method also includes generating ecological restoration plans based on comprehensive assessment reports of shelterbelt degradation. According to the degradation level distribution map of different regions, a pre-set restoration measure library is matched to obtain a preliminary set of restoration measures. Based on the degradation acceleration time window in the degradation state prediction map, the start time and duration of implementation are planned for each measure in the preliminary set of restoration measures. Combining soil property data and microclimate parameter sequences, the measure parameters in the preliminary set of restoration measures are localized and optimized. These parameters include vegetation density, irrigation quota, and fertilizer application rate. The output is an ecological restoration plan containing the optimized measures, implementation schedule, and parameter configurations.

[0103] In its implementation, the comprehensive assessment report generation module integrates the degradation level determination results and degradation trend vectors from the intelligent degradation identification module. The degradation level determination result is a data layer containing degradation level classifications for different spatial locations, and the degradation trend vector is a mathematical vector containing information on the future direction and magnitude of changes in various features. The comprehensive assessment report generation module spatially correlates the degradation level determination results with the administrative division layer and sub-block layer in the geographic information system. Through spatial overlay analysis and attribute connection operations, it adds corresponding administrative affiliation information and forest tenure information to each degradation level patch, generating a degradation level distribution map with administrative affiliation and forest tenure information. In some embodiments, the degradation trend vector is discretized on the time axis to generate a degradation state prediction map set for multiple future time nodes. The degradation trend vector is input into a state transition prediction model. Based on the current degradation level distribution and the direction and rate of change indicated in the degradation trend vector, the state transition prediction model iteratively calculates the degradation state of each spatial location in the future according to a set time step. The time step can be set to one year or one growing season. The iterative calculation generates a spatial distribution map of the degradation state for multiple time nodes such as the fifth year and the tenth year in the future. These prediction maps constitute a degradation state prediction map set.

[0104] Optionally, the boundaries and attribute information of degraded areas requiring intervention are extracted from the degradation level distribution map. A threshold condition for requiring intervention is set as continuous map patches reaching "severe degradation" or "moderate degradation" and with an area greater than a certain threshold. Using the spatial query and filtering functions of the geographic information system, polygonal boundaries of degraded areas meeting the conditions are extracted, along with their attribute information, including area, average degradation level, administrative division, and forest tenure owner. Degradation acceleration time windows are extracted from the degradation state prediction map set. The changes in degradation level at the same spatial location in prediction maps at different future time points are analyzed. Time periods of rapid or continuous deterioration in degradation level are identified as degradation acceleration time windows. The boundaries and attribute information of degraded areas are combined and correlated with degradation acceleration time windows to constitute the core content of the comprehensive assessment report on protective forest degradation. The core content of the comprehensive assessment report is organized in the form of text, data tables, and thematic maps.

[0105] It is understandable that an ecological restoration plan is generated based on the comprehensive assessment report of protective forest degradation. According to the degradation level distribution map of different regions, a pre-set restoration measure library is matched. This library stores restoration measures corresponding to different degradation levels. A preliminary set of restoration measures for each region in the map is obtained through querying and matching. Based on the degradation acceleration time window in the degradation state prediction map, the start time and duration of each measure in the preliminary restoration measure set are planned. For regions near the degradation acceleration time window, the start time is set to the current or next suitable season. For regions far from the degradation acceleration time window, the start time can be appropriately delayed but must be before the start of the time window. The duration of the measure is determined based on the type and intensity of the matched measure. Combining soil property data and microclimate parameter sequences, the measure parameters in the preliminary restoration measure set are localized and optimized. These parameters include vegetation density, irrigation quota, and fertilizer application rate. The optimization formula for vegetation density is:

[0106]

[0107] in: This represents the optimized vegetation density. Represents the basic vegetation configuration density in the restoration measures library. It is the adjustment coefficient. It refers to the organic matter content of the local soil. This represents the average soil organic matter content in the region. Irrigation quotas are adjusted based on the difference between the local average annual precipitation and potential evapotranspiration, while fertilizer application is adjusted according to soil nutrient test results. In implementation, an ecological restoration plan is output, including optimized measures, an implementation timeline, and parameter configurations. The plan is presented in structured documents and charts, clearly outlining the geographical scope of each area requiring restoration, the optimized combination of restoration measures, the specific parameters of each measure, the start and end dates of the plan, and the expected resource input.

[0108] See Figure 5This is a time-series heatmap predicting the degradation levels of different areas of shelterbelts, showcasing the current, 5-year, and 10-year trends in degradation levels for each area. Areas A (currently 2, 10-year target 4), E (currently 3, 10-year target 4), and F (currently 2, 10-year target 4) are projected to upgrade from mild / moderate to extremely severe degradation within 10 years. Areas D and G (currently at 4, expected to remain extremely severe in the next 10 years) are key areas for long-term intervention. Areas C (currently 1, 10-year target 3) and H (currently 1, 10-year target 3) show relatively slower degradation, but still require proactive prevention and control. This map is used for time-series planning of shelterbelt ecological restoration: by predicting the time-series degradation levels of each area, short-term intervention plans can be prioritized for rapidly deteriorating areas (such as A, E, and F), while long-term restoration strategies can be implemented for persistently extremely severe areas (D and G), improving the accuracy and timeliness of ecological restoration.

[0109] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0110] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A protective forest monitoring and assessment system based on intelligent visual recognition, characterized in that, Includes the following modules: The multi-source information synchronous acquisition module is used to acquire multi-source synchronous sensing information for the target shelterbelt. The multi-source synchronous sensing information is captured by a visual sensing network, a terrain scanning device, and a meteorological monitoring unit at the same time coordinate. The multi-dimensional feature extraction module is used to perform multi-dimensional feature extraction operations on the multi-source synchronous sensing information to obtain the vegetation growth spatial field, which includes the canopy morphology topology, canopy closure change history and biomass distribution matrix. The degradation map construction module is used to construct a degradation feature evolution map based on the vegetation growth spatial field. The degradation feature evolution map includes the outline of the degradation patch, the degradation spread trajectory, and the degradation intensity distribution heat map. The intelligent degradation identification module integrates a trained multi-level degradation identification network to perform deep feature analysis on the degradation feature evolution map, and obtain degradation level determination results and degradation trend vectors. The comprehensive assessment report generation module is used to integrate the degradation level determination results and the degradation trend vector to form a comprehensive assessment report on the degradation of protective forests that includes spatial location information and time series information; The degradation map construction module constructs a degradation feature evolution map based on the vegetation growth spatial field, including the following steps: Regions with missing canopy coverage and regions with abnormally decreased canopy height were identified in the canopy morphology topology, and the identified regions were marked as initial degraded plaques; The process of canopy density change is analyzed, the canopy density decay rate and decay direction are calculated, and the degradation and spread trajectory is simulated based on the decay rate and decay direction. Locate regions with significantly low biomass values ​​in the biomass distribution matrix, calculate the biomass gradient difference between the regions with significantly low biomass values ​​and the surrounding regions, and generate the degradation intensity distribution heatmap based on the biomass gradient difference. The spatial coordinates of the initial degraded patch, the simulated path of the degradation spread trajectory, and the intensity values ​​of the degradation intensity distribution heatmap are layered and correlated to form the degradation feature evolution map.

2. The protective forest monitoring and evaluation system based on intelligent visual recognition according to claim 1, characterized in that, The multi-dimensional feature extraction module performs multi-dimensional feature extraction operations on the multi-source synchronous sensing information to obtain the vegetation growth spatial field, including the following operations: The multi-source synchronous sensing information is aligned and gridded in the spatiotemporal dimension to generate a fusion data cube with a unified spatial reference and timestamp. The spectral reflectance sequence, point cloud high-order sequence, and microclimate parameter sequence are separated from the fused data cube; The spectral reflectance sequence was used to perform vegetation index inversion calculations to generate leaf area index distribution maps, chlorophyll content distribution maps, and water stress index distribution maps. The point cloud elevation sequence is reconstructed and segmented in three dimensions to generate a digital elevation model, a canopy height model, and individual tree segmentation boundaries; By integrating the leaf area index distribution map, the digital elevation model, the canopy height model, and the microclimate parameter sequence, the canopy morphology topology, the canopy closure change history, and the biomass distribution matrix are constructed through spatial interpolation and weighted calculation.

3. The protective forest monitoring and evaluation system based on intelligent visual recognition according to claim 2, characterized in that, The process of performing three-dimensional reconstruction and segmentation on the point cloud elevation sequence to generate a digital elevation model, a canopy height model, and individual tree segmentation boundaries includes the following steps: The point cloud high-order sequence is subjected to denoising and filtering to remove outliers and noisy data, resulting in clean point cloud data; Based on the clean point cloud data, a three-dimensional surface model is constructed using a triangulation algorithm. The three-dimensional surface model represents the continuous surface of the terrain and canopy. Ground point cloud is extracted from the three-dimensional surface model, and a digital elevation model is generated using an interpolation method. Non-ground point clouds are separated from the three-dimensional surface model, and the elevation values ​​of the corresponding locations in the digital elevation model are subtracted from the elevation values ​​of the non-ground point clouds to obtain normalized canopy height point clouds. The canopy height point cloud is clustered and segmented, and the boundaries of individual tree canopies are identified based on the spatial distance and density characteristics of the point cloud to generate individual tree segmentation boundaries; Based on the canopy height point cloud and the single tree segmentation boundary, a canopy height model is generated by statistically analyzing the point cloud height distribution within each segmentation region.

4. The protective forest monitoring and evaluation system based on intelligent visual recognition according to claim 3, characterized in that, The analysis of the canopy closure change process, calculation of the canopy closure decay rate and decay direction, and simulation of the degradation spread trajectory based on the decay rate and decay direction include the following steps: Extract canopy closure raster data from multiple time points from the canopy closure change history, calculate the difference in canopy closure values ​​between adjacent time points, and obtain the canopy closure change raster. Trend analysis was performed on the canopy density change grid. The slope of the canopy density change of each grid cell over time was fitted using a linear regression method, and the slope of the change was used as the canopy density decay rate. Based on the spatial distribution of canopy closure decay rate, the rate gradient of the surrounding area of ​​each grid cell is calculated, and the direction of the fastest rate decrease is determined as the canopy closure decay direction. Centered on the initial degraded plaque, a cellular automata model is used to simulate the expansion path of the degraded plaque based on the canopy closure decay rate and canopy closure decay direction. The expansion path takes into account the continuity of rate magnitude and direction. The simulated expansion path is verified and adjusted against historical degraded patches to ultimately generate a degradation spread trajectory.

5. The protective forest monitoring and evaluation system based on intelligent visual recognition according to claim 4, characterized in that, The intelligent degradation identification module obtains the degradation level determination result and degradation trend vector, including the following steps: The multi-level degradation identification network performs shallow feature extraction on the degradation feature evolution map to obtain the shape complexity, edge fragmentation and spatial clustering of the degradation patches; The multi-level degradation identification network extracts mid-level features from the degradation propagation trajectory to obtain the trajectory's tortuosity, expansion speed, and directional stability. The multi-level degradation identification network performs deep feature extraction on the degradation intensity distribution heatmap to obtain the centrality, heterogeneity, and spatial autocorrelation of intensity changes; The shape complexity, the expansion speed, and the centrality of the intensity change are fused and mapped to output the degradation level determination result. The spatial clustering, directional stability, and spatial autocorrelation are extrapolated over time to output the degradation trend vector describing the future direction of change.

6. The protective forest monitoring and evaluation system based on intelligent visual recognition according to claim 5, characterized in that, The system also includes multi-scale environmental compensation for the degradation level determination results, including: Historical climate sequence and soil property data of the target shelterbelt area are obtained, and the environmental carrying capacity is calculated. At the landscape scale, the overall degradation index in the degradation level determination result is compensated and corrected based on the environmental carrying capacity. At the patch scale, the degradation level of individual degraded patches is adjusted differentially based on the soil attribute data, with the degradation level of areas with lower soil fertility being increased and the degradation level of areas with higher soil fertility being decreased. At the pixel scale, the local intensity values ​​in the degradation intensity distribution heatmap are normalized for illumination conditions by incorporating the terrain shadow coefficient. Output the final degradation level after multi-scale environmental compensation.

7. The protective forest monitoring and evaluation system based on intelligent visual recognition according to claim 6, characterized in that, At the landscape scale, the overall degradation index in the degradation level determination result is compensated and corrected based on the environmental carrying capacity, specifically as follows: The ratio of the environmental carrying capacity to the standard reference capacity is calculated to obtain the environmental adjustment coefficient; Multiply the overall degradation index by the environmental adjustment coefficient to obtain the environmentally corrected overall degradation index; When the environmental carrying capacity is lower than a preset threshold, the drought stress enhancement module is activated. The drought stress enhancement module further corrects the overall degradation index after environmental correction based on the water deficit.

8. The protective forest monitoring and evaluation system based on intelligent visual recognition according to claim 7, characterized in that, The comprehensive assessment report generation module integrates the degradation level determination results and the degradation trend vector to form a comprehensive assessment report on the degradation of protective forests that includes spatial location information and time series information, including the following steps: The degradation level determination results are spatially correlated with the administrative division layer and small plot layer in the geographic information system to generate a degradation level distribution map with administrative affiliation and forest tenure information; Discretize the degradation trend vector on the time axis to generate a set of degradation state prediction maps for multiple future time nodes; Extract the boundaries and attribute information of the degradation areas that urgently need intervention from the degradation level distribution map, and extract the time window of degradation acceleration from the degradation state prediction map set; The core content of the comprehensive assessment report on the degradation of protective forests is formed by combining the boundaries of the degraded areas that urgently require intervention, the attribute information, and the time window of accelerated degradation.

9. The protective forest monitoring and evaluation system based on intelligent visual recognition according to claim 8, characterized in that, The system also includes generating an ecological restoration plan based on the comprehensive assessment report of shelterbelt degradation, including: Based on the degradation level of different regions in the degradation level distribution map, a preset set of restoration measures is obtained by matching them with the preset restoration measures library. Based on the degradation acceleration time window in the degradation state prediction map set, the start time and duration of implementation are planned for each measure in the preliminary recovery measure set; Combining the soil property data and the microclimate parameter sequence, the parameters of the measures in the preliminary restoration measure set are optimized for localization. The measures parameters include vegetation configuration density, irrigation quota and fertilizer application amount. The output includes an ecological restoration plan with optimized measures, implementation timeline, and parameter configuration.

Images