An ecological protection red line vegetation anomaly rapid monitoring method

Abstract

This application relates to the field of ecological monitoring technology and discloses a method for rapid monitoring of vegetation anomalies along ecological protection red lines. The method includes the following steps: acquiring remote sensing images to form a pixel matrix; extracting the single-dimensional feature of vegetation cover from each pixel, calculating the negative deviation, and comparing it against a threshold to filter a first-level suspected set; calculating the leaf area index and total primary productivity for the pixels in this set, and constructing a three-dimensional feature vector together with the vegetation cover; constructing an asymmetric weight matrix based on this vector to solve for the weighted Mahalanobis distance squared term, and filtering a second-level candidate anomaly set; searching for baseline control points within the spatial neighborhood of pixels in the second-level candidate anomaly set, and generating a confirmed anomaly pixel set by executing branch routing logic based on the total number of baseline control points; performing spatial morphological correction and raster vectorization on the confirmed anomaly pixel set, and outputting a map of the area of ​​abnormal vegetation changes along the ecological protection red line. This invention reduces computational load, eliminates interference from regional meteorological fluctuations, and improves monitoring accuracy.

Description

Technical Field

[0001] This invention relates to the field of ecological monitoring technology, specifically a method for rapid monitoring of vegetation anomalies along ecological protection red lines. Background Technology

[0002] Ecological protection red lines are important areas for maintaining ecological security. Continuous spatial monitoring of vegetation cover and growth within the red lines is a routine part of ecological protection work. Currently, using optical remote sensing images to extract ecological parameters such as vegetation cover and leaf area index for change detection is the main way to achieve vegetation anomaly monitoring.

[0003] When dealing with large spatial areas, existing monitoring methods directly perform joint extraction of multi-dimensional ecological features and complex distance calculations on all spatial pixels, which increases the computational resource load and reduces data processing time. In the anomaly detection stage, existing algorithms are easily affected by regional climate fluctuations. When there is a large-scale reduction in precipitation or abnormal temperature, vegetation parameters in the region will generally decline. Conventional methods are unable to distinguish between normal natural environmental background fluctuations and local real degradation signals, which can easily lead to false alarms.

[0004] In the joint calculation of multiple indicators, existing technologies typically apply equal weights to various ecological parameters, failing to consider the differences in stability of different parameters over long-term historical observations. Because the natural fluctuation patterns of each parameter differ, parameters with poor stability introduce noise interference into the calculation, masking actual vegetation anomalies and leading to inaccurate monitoring results. Therefore, this invention proposes a rapid monitoring method for vegetation anomalies along ecological protection red lines to address the shortcomings of existing technologies. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a rapid monitoring method for vegetation anomalies along ecological protection red lines. This method solves the problems of excessive computational load and low processing time in the calculation of multi-dimensional parameters across the entire region, the inability to distinguish between regional meteorological fluctuations and local real degradation signals leading to false alarms, and the failure to consider the historical stability differences of ecological parameters in the calculation of multiple indicators, which introduces noise interference and results in inaccurate monitoring results.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] A method for rapid monitoring of vegetation anomalies along ecological protection red lines includes the following steps:

[0008] Acquire optical remote sensing image data and process it to form a pixel matrix composed of multiple spatial pixels;

[0009] For each spatial pixel in the pixel matrix, the normalized relative density of vegetation cover is calculated as a one-dimensional feature, and the one-dimensional negative deviation is calculated based on the one-dimensional feature. The one-dimensional negative deviation is compared with the preset first-order truncation lower limit threshold to filter and generate a first-level suspected set.

[0010] For the pixels in the first-level suspected set, calculate the normalized relative density of leaf area index and the normalized relative density of total primary productivity, and construct a three-dimensional normalized feature vector together with the normalized relative density of vegetation cover.

[0011] For the pixels in the first-level suspected set, based on the three-dimensional normalized feature vector, an asymmetric diagonal weight matrix is ​​constructed according to the historical steady-state variation coefficient and the weighted Mahalanobis distance squared term is solved to filter pixels that exceed the preset chi-square distribution distance critical threshold and generate a second-level candidate anomaly set.

[0012] Within the spatial neighborhood of a pixel in the secondary candidate anomaly set, homologous normal background pixels are searched as background control points, and the total number of background control points is counted.

[0013] Based on whether the total number of the baseline control points reaches the preset effective number control line, the branch routing logic is executed to generate a set of confirmed abnormal pixels.

[0014] Spatial morphological correction and raster vectorization transformation are performed on the confirmed abnormal pixel set to output a map of the area of ​​abnormal vegetation change along the ecological protection red line.

[0015] Preferably, constructing the three-dimensional normalized feature vector of a spatial pixel specifically includes:

[0016] Based on ecological function zoning data and land cover classification raster data, each spatial cell in the spatial cell matrix is ​​assigned a spatial assessment unit attribute label containing ecological function zoning index and vegetation type index.

[0017] For each spatial pixel, the evaluation benchmark reference values ​​for vegetation cover, leaf area index, and total primary productivity are obtained. The evaluation benchmark reference values ​​are obtained using an open mechanism: if there are publicly available benchmark value results for the area where the target pixel is located, the publicly available benchmark value results are directly extracted as the evaluation benchmark reference values; if there are no publicly available benchmark value results, then, with the target spatial pixel as the center, neighboring pixels of the same spatial evaluation unit attribute label within a certain spatial range are selected, and the historical maximum values ​​of vegetation cover, leaf area index, and total primary productivity of the neighboring pixels within the historical observation period are extracted, and their average values ​​are calculated and set as the evaluation benchmark reference values.

[0018] The relative density of spatial parameters is obtained by calculating the division ratio of the observed vegetation cover, leaf area index, and total primary productivity parameters of the current observation phase corresponding to the optical remote sensing image data with the corresponding evaluation benchmark reference values.

[0019] The extreme value normalization mathematical method is used to transform the relative density of spatial parameters into a numerical distribution range of zero to one, thereby generating normalized relative density values.

[0020] The normalized relative density of vegetation cover, leaf area index, and total primary productivity of a single spatial pixel are linearly concatenated to form a three-dimensional normalized feature vector. Through these steps, the differences in the absolute values ​​of basic parameters among different vegetation categories and ecological functional zones can be eliminated, establishing a unified benchmark for anomaly measurement.

[0021] Preferably, the process of generating a first-level suspected set by comparing the one-dimensional negative deviation with a preset first-order truncation lower limit specifically includes:

[0022] Historical optical remote sensing image data of the same phenological period as the spatial pixel are extracted, and the vegetation cover is normalized to the historical relative density data within the historical time series smoothing window. The historical average value of the historical data is calculated to obtain the historical average value of the vegetation cover feature.

[0023] Calculate the one-dimensional negative deviation of the normalized relative density of vegetation cover from the historical average value of the vegetation cover feature at the current observation time.

[0024] A trigger discriminator function is constructed to compare the single-dimensional negative deviation with the first-order truncation lower limit threshold. Based on the output value of the trigger discriminator function, the spatial coordinate system attributes of spatial pixels are collected and aggregated to generate the first-level suspected set. Through the above steps, the single-dimensional vegetation cover feature with lower extraction computational requirements is truncated, reducing the high-dimensional parameter calculations in areas without degradation risk and completing the spatial dimensionality reduction operation.

[0025] Preferably, constructing an asymmetric diagonal weight matrix based on historical steady-state variation coefficients specifically includes:

[0026] The historical steady-state variation coefficient of the corresponding ecological parameter is obtained by calculating the ratio of the historical standard deviation to the historical mathematical mean of each ecological parameter in the time series after invalid data removal for spatial pixels.

[0027] When the extracted historical steady-state coefficient of variation is less than the set minimum positive real constant, the historical steady-state coefficient of variation is equivalently replaced with the minimum positive real constant.

[0028] Calculate the reciprocal of the historical steady-state coefficient of variation to obtain the asymmetric weighting factor;

[0029] By placing asymmetric weight factors on the main diagonal and setting the elements on the off-diagonal positions to zero, an asymmetric diagonal weight matrix in the form of a third-order square matrix is ​​constructed.

[0030] Preferably, based on the three-dimensional normalized feature vector, an asymmetric diagonal weight matrix is ​​constructed according to the historical steady-state variation coefficient, and the weighted Mahalanobis distance squared term is solved. Pixels exceeding the preset chi-square distribution distance critical threshold are screened to generate a secondary candidate anomaly set, specifically including:

[0031] Based on the time series of three-dimensional normalized feature vectors within the historical observation period, the third-order historical covariance matrix of a single spatial pixel is calculated.

[0032] A minimal positive real regularization parameter is added to the main diagonal elements of the third-order history covariance matrix to generate a regularized covariance matrix.

[0033] Read the current period three-dimensional normalized feature vector, three-dimensional mean vector, the asymmetric diagonal weight matrix, and the inverse matrix of the regularized covariance matrix of the current observation phase, and calculate the weighted Mahalanobis distance squared term value of the spatial pixel.

[0034] If the weighted Mahalanobis distance squared term exceeds the chi-square distribution distance critical threshold, spatial pixels are filtered and uniformly classified into the secondary candidate anomaly set. Through the above steps, the proportion of historically stable ecological parameter dimensions in distance calculation can be adjusted using an asymmetric diagonal weight matrix, reducing the numerical interference of larger natural noise dimensions, and avoiding matrix divergence caused by collinearity among ecological parameters by adding regularization parameters.

[0035] Preferably, the execution of branch routing logic to generate a set of confirmed abnormal pixels specifically includes:

[0036] When the total number of background control points reaches the preset effective number control line, extract the weighted Mahalanobis distance squared value of all background control points associated with a specific spatial cell, calculate the arithmetic mean to generate the average distance of background fluctuation in the region.

[0037] The net anomaly deviation is obtained by subtracting the average distance of regional background fluctuations from the weighted squared Mahalanobis distance of a specific spatial cell. Spatial cells with a net anomaly deviation greater than the preset net anomaly control threshold are loaded into the confirmed anomaly cell set.

[0038] If the total number of baseline control points does not reach the preset effective number control line, pixels in the secondary candidate anomaly set are directly loaded into the confirmed anomaly pixel set. Through the above steps, the phenomenon of general natural parameter decline caused by regional meteorological fluctuations can be eliminated, and the anomalous signal can be preserved in the case of missing homologous normal reference base.

[0039] Preferably, when the total number of baseline control points reaches the preset effective retrieval control line, generating the set of confirmed abnormal pixels further includes:

[0040] After loading spatial cells with net anomaly deviation greater than the net anomaly control threshold into the confirmed anomaly cell set, spatial connectivity calculation is performed on the confirmed anomaly cell set, and isolated cells with a number of connected cells less than the minimum anomaly contiguous area threshold are deleted.

[0041] Preferably, the spatial morphological correction and raster vectorization transformation of the confirmed anomalous pixel set specifically includes:

[0042] A two-dimensional binary image matrix is ​​established, and the spatial pixel positions within the confirmed anomalous pixel set are assigned a value of one, while the remaining positions are assigned a value of zero.

[0043] By pre-setting morphological structural elements, the two-dimensional binary image matrix is ​​first dilated and then eroded to form a mathematical morphological closing operation, resulting in a modified two-dimensional binary image matrix.

[0044] Perform raster vectorization transformation on the modified two-dimensional binary image matrix to form a vector polygon closed boundary.

[0045] Preferably, the closed boundary of the vector polygon is the vector polygon of the current period's anomalous patches. The rapid monitoring method for vegetation anomalies along the ecological protection red line also includes:

[0046] Connect to a historical spatial database to extract vector polygons of historical anomalous patches;

[0047] Perform polygon topology overlay calculation on the vector polygons of current period abnormal features and vector polygons of historical period abnormal features to calculate the area of ​​newly added abnormal regions and the area of ​​persistent abnormal regions.

[0048] Preferably, the output map of areas with abnormal vegetation changes along the ecological protection red line specifically includes:

[0049] By integrating the spatial location information of the vector polygons of current anomalous patches, the area of ​​newly added anomalous areas, the area of ​​persistent anomalous areas, and the associated net anomalous deviation attribute, the spatiotemporal evolution results of ecological anomalies are generated.

[0050] The spatiotemporal evolution results of ecological anomalies are packaged into an ecological violation early warning report, and the vector polygons of the current abnormal patches are overlaid on the spatial geographic reference base map for output.

[0051] This invention provides a method for rapid monitoring of vegetation anomalies along ecological protection red lines. It has the following beneficial effects:

[0052] 1. This invention employs a two-stage screening process that combines initial screening with detailed investigation of multidimensional features. The rapid calculation of the negative deviation of vegetation cover in one dimension can first eliminate most areas without anomalies, concentrate computational resources on a few suspected abnormal pixels, and then perform complex multidimensional weighted Mahalanobis distance calculation only on the first-level suspected set. The hierarchical processing mechanism reduces the overall computational load and improves the monitoring efficiency and timeliness of large-scale ecological protection red line areas.

[0053] 2. This invention constructs an asymmetric diagonal weight matrix based on historical steady-state variation coefficients. When solving the weighted Mahalanobis distance, the asymmetric diagonal weight matrix can amplify the contribution of ecological parameter dimensions with stable historical performance and low variability in anomaly detection, while suppressing noise interference from parameter dimensions with large historical fluctuations. The adaptive weighting method enables the monitoring model to better distinguish between real vegetation degradation signals and normal natural fluctuations, thereby improving the accuracy and reliability of anomaly identification.

[0054] 3. This invention introduces a background control point correction logic based on spatial neighborhood. By searching for homologous normal background pixels around suspected anomalous pixels and calculating and deducting the influence of regional background fluctuations, it can effectively isolate the general vegetation decline caused by regional meteorological factors such as drought and low temperature. This design can distinguish between real, local anomalous changes and large-scale natural environmental stresses, reduce the false alarm rate caused by natural factors, and make the final output anomalous patches more accurately point to the areas of interest. Attached Figure Description

[0055] Figure 1 This is a schematic diagram of the system architecture of the present invention;

[0056] Figure 2 This is a schematic diagram of the overall process of the method of the present invention;

[0057] Figure 3 This is a schematic diagram comparing the system performance improvement effects of the present invention;

[0058] Figure 4 This is a schematic diagram comparing the extraction of multidimensional normalized features of temporal ecological parameters in this invention.

[0059] Figure 5 This is a schematic diagram illustrating the background stripping and denoising and high-dimensional anomaly confirmation determination of the present invention. Detailed Implementation

[0060] The technical solutions in 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.

[0061] Please see Figure 1 This invention provides a rapid monitoring system for abnormal vegetation along ecological protection red lines, comprising: a data preprocessing module, a preliminary screening triggering module, a high-dimensional verification module, a background stripping module, and a map generation module.

[0062] The data preprocessing module communicates with the initial screening trigger module and the high-dimensional verification module. The data preprocessing module outputs single-dimensional feature data of vegetation coverage to the initial screening trigger module, and outputs a three-dimensional normalized feature vector containing spatial pixels to the high-dimensional verification module.

[0063] The initial screening trigger module communicates with the high-dimensional inspection module and the background stripping module. The initial screening trigger module outputs a first-level suspected set containing spatial pixel coordinates to the high-dimensional inspection module, and outputs the coordinates of normal spatial pixels that are not included in the first-level suspected set to the background stripping module.

[0064] The high-dimensional inspection module communicates with the background stripping module and outputs a set of secondary candidate anomalies containing pixel coordinates and associated weighted Mahalanobis distance data to the background stripping module.

[0065] The background stripping module communicates with the image generation module and outputs a set of confirmed abnormal pixels generated by background verification to the image generation module.

[0066] See attached document Figure 2 This invention provides a method for rapid monitoring of vegetation anomalies along ecological protection red lines, comprising the following steps:

[0067] Optical remote sensing image data is acquired and radiometric calibration and atmospheric correction are performed. Spatial assessment units are divided based on ecological functional zoning data, and benchmark reference values ​​are obtained. The design of the benchmark reference values ​​is open: if there are legal or suitable publicly available benchmark values ​​for the target area, the existing publicly available benchmark value results are directly adopted as the benchmark reference values; if there are no publicly available benchmark values, the average of the historical highest values ​​of vegetation cover, leaf area index, and total primary productivity of the same type of vegetation pixels within a certain spatial range around the target pixel during the historical observation period is selected as the benchmark reference value. The extreme values ​​of vegetation cover, leaf area index, and total primary productivity for each vegetation type during the historical observation period are extracted as benchmark reference values. The vegetation cover parameters, leaf area index parameters, and total primary productivity parameters of the currently input remote sensing image are extracted. The spatial parameter relative density of the remote sensing image pixels is calculated and extreme value normalization is performed. The three-dimensional normalized feature vector of the spatial pixels is constructed using the normalization processing results.

[0068] The system independently extracts one-dimensional features of vegetation cover across the entire spatial domain, calculates the negative deviation of the current vegetation cover of remote sensing image pixels from the historical average vegetation cover, performs numerical comparison by setting a first-order truncation lower limit threshold, blocks the subsequent calculation path of spatial pixels whose negative deviation does not reach the first-order truncation lower limit threshold, collects the spatial coordinate system set of spatial pixels whose negative deviation exceeds the first-order truncation lower limit threshold, and generates a first-level suspected set.

[0069] The system receives the 3D normalized feature vectors of the pixels contained in the first-level suspected set, extracts the historical steady-state variation coefficients of the pixels in the first-level suspected set, calculates the reciprocal of the variation coefficients based on the historical steady-state variation coefficients, constructs an asymmetric diagonal weight matrix, introduces a minimum positive real number regularization parameter into the historical time-series feature covariance matrix, generates a regularized covariance matrix, combines the asymmetric diagonal weight matrix and the regularized covariance matrix to solve the weighted Mahalanobis distance, compares the weighted Mahalanobis distance with the set chi-square distribution distance critical threshold, and filters out pixels whose weighted Mahalanobis distance exceeds the chi-square distribution distance critical threshold and classifies them into the second-level candidate anomaly set.

[0070] The background stripping module reads the spatial coordinate attributes of pixels in the secondary candidate anomaly set, searches for spatially neighboring pixels that belong to the same ecological functional zoning and the same vegetation type and are not included in the primary suspected set, sets pixels that meet the search conditions as baseline control points, and counts the total number of baseline control points.

[0071] If the total number of determined baseline control points reaches the preset effective number control line, calculate the weighted Mahalanobis distance arithmetic mean of the determined baseline control points, set the weighted Mahalanobis distance arithmetic mean as the regional background fluctuation mean distance, calculate the numerical difference between the weighted Mahalanobis distance and the regional background fluctuation mean distance of the pixels in the secondary candidate anomaly set, obtain the net anomaly deviation, retain the pixels with a net anomaly deviation greater than zero and remove the remaining pixels, perform spatial connectivity calculation operation on the retained pixels, delete isolated pixels whose number of connected pixels does not reach the threshold, and generate a set of confirmed anomaly pixels;

[0072] If the total number of determined baseline control points does not reach the preset effective number control line, the calculation of the average distance of regional background fluctuations is stopped, and the pixels in the secondary candidate anomaly set are directly loaded into the confirmed anomaly pixel set.

[0073] A two-dimensional binary image matrix is ​​established by receiving a set of confirmed abnormal pixels. Mathematical morphology closing operations are used to eliminate internal holes and edge jaggedness. Edge features are extracted to form vector polygon closed boundaries. The vector polygon closed boundaries are superimposed on the spatial geographic reference base map to output a map of the area of ​​abnormal vegetation change along the ecological protection red line.

[0074] See attached document Figure 2 In this embodiment of the invention, the specific steps of data preprocessing and feature labeling performed by the data preprocessing module are as follows:

[0075] The data preprocessing module acquires time-series optical remote sensing image data covering the ecological protection red line area. For the radiometric calibration, atmospheric correction and geometric correction operations of the optical remote sensing image data, those skilled in the art can use existing remote sensing image processing platforms or open source algorithm libraries. The radiometric calibration and atmospheric correction of optical remote sensing image data are well known technologies in this field and will not be described in detail here. The preprocessed optical remote sensing image data is mapped to a unified spatial projection coordinate system to form a pixel matrix composed of multiple spatial pixels.

[0076] The data preprocessing module reads the ecological functional zoning vector data and the land cover classification raster data, aligns the ecological functional zoning vector data with the spatial projection coordinate system, and assigns a spatial assessment unit attribute label to each spatial pixel in the pixel matrix. The spatial assessment unit attribute label includes the ecological functional zoning index and the vegetation type index. The ecological functional zoning index refers to the specific ecological functional zone in which the spatial pixel is located, and the vegetation type index refers to the natural vegetation cover category to which the spatial pixel belongs. The specific sub-feature categories of the vegetation type index include forest ecosystem category, shrub ecosystem category, grassland ecosystem category and farmland ecosystem category.

[0077] The data preprocessing module extracts the historical maximum values ​​of vegetation cover, leaf area index, and total primary productivity for each spatial assessment unit attribute label within the historical observation period. Here, the historical observation period is usually selected as a complete data smoothing window of 3-5 years before the current monitoring phase to eliminate the interference caused by extreme climate fluctuations in a single year on the benchmark reference value. For the calculation model of vegetation cover, the inversion model of leaf area index, and the estimation model of total primary productivity, those skilled in the art can select the corresponding empirical model or physical process model according to the specific remote sensing sensor type. The data preprocessing module uniformly sets the extracted public benchmark value or the average value of the historical highest value of pixels within a certain spatial range as the evaluation benchmark reference value.

[0078] The data preprocessing module calculates the relative density of spatial parameters for each spatial cell based on the evaluation benchmark reference value. Because the absolute values ​​of basic ecological parameters vary significantly among different vegetation ecosystem types, such as forests and grasslands, and different ecological functional zones, directly using absolute raw observation values ​​cannot provide a unified anomaly measurement across regions and types. Therefore, by calculating the ratio of the current observation value to the benchmark value under the historical optimal state, the inherent differences in vegetation density can be eliminated, purely reflecting the current growth status of the vegetation. The data preprocessing module calculates the division ratio of the vegetation cover observation value of a specific spatial cell at the current observation time to the corresponding evaluation benchmark reference value, obtaining the relative density of vegetation cover. For the leaf area index (LAI) and total primary productivity (TPP) parameters, the data preprocessing module performs the same division ratio calculation, obtaining the relative density of LAI and TPP respectively. The calculation formulas for the relative density of spatial parameters are as follows:

[0079] ;

[0080] in, This represents the spatial coordinate position of a specific spatial cell; Represents the phase variable of the observation; Dimensions representing ecological parameters The constant values ​​are 1, 2 or 3, corresponding to the vegetation cover parameter, leaf area index parameter and total primary productivity parameter, respectively. Spatial coordinate position of a specific spatial cell During the observation phase The The relative density values ​​of spatial parameters in each dimension; Spatial coordinate position of a specific spatial cell During the observation phase The The original absolute values ​​of ecological parameters in each dimension were observed; Spatial coordinate position of a specific spatial cell The assigned ecological functional zoning index variables; Spatial coordinate position of a specific spatial cell The vegetation type index variable assigned to it; This indicates that a specific spatial cell is located within a given ecological functional zoning area. With vegetation type Obtained under combined conditions, the first Benchmark values ​​for ecological parameters in each dimension.

[0081] The data preprocessing module employs an extreme value normalization mathematical method to transform the calculated spatial parameter relative density into a numerical distribution range between zero and one, generating normalized relative density values. The data preprocessing module targets specific dimensions. Extracting the same ecological functional zoning within the historical observation period With vegetation type index The historical maximum and minimum values ​​of the relative density of spatial parameters under the combined conditions are used as boundary parameters for extreme value normalization. The normalization formula is as follows:

[0082] ;

[0083] in, This represents the normalized relative density value after extreme value normalization. Indicates the ecological functional zoning With vegetation type Under combined conditions, the first The historical statistical minimum values ​​of the relative density of spatial parameters in each dimension; Indicates the ecological functional zoning With vegetation type Under combined conditions, the first The historical maximum values ​​of the relative density of spatial parameters in each dimension.

[0084] The data preprocessing module linearly concatenates the normalized relative density of vegetation cover, normalized relative density of leaf area index, and normalized relative density of total primary productivity of a single spatial pixel according to the order of the ecological parameter dimension identifiers, forming a three-dimensional normalized feature vector. The formula for constructing the three-dimensional normalized feature vector is as follows:

[0085] ;

[0086] in, Represents spatial coordinate position quantity During the observation phase The generated three-dimensional normalized eigenvector column matrix; Represents spatial coordinate position quantity During the observation phase The normalized relative density scalar value of vegetation cover; Represents spatial coordinate position quantity During the observation phase The leaf area index normalized relative density scalar value; Represents spatial coordinate position quantity During the observation phase The normalized relative density scalar value of total primary productivity; To represent the matrix transpose operation, and to prevent division by zero errors in normalization calculations caused by the historical maximum and minimum values ​​of the spatial parameter relative density being equal, a very small positive real number (e.g., 10) is superimposed after the denominator term. -6 ) is used as a regularization protection term.

[0087] The three-dimensional normalized feature vectors completely record the set of comprehensive ecological representation parameters of a single spatial pixel under a unified mathematical dimension. The data preprocessing module outputs all the three-dimensional normalized feature vectors to the downstream processing module.

[0088] See attached document Figure 2 In this embodiment of the invention, the specific steps of the initial screening trigger module in performing computational power dimensionality reduction and initial screening triggering are as follows:

[0089] Since the high-dimensional model inversion process of leaf area index (LAI) and total primary productivity (TPP) parameters requires a large amount of computing power, performing full-domain synchronous high-dimensional calculations in the vast ecological protection redline area would waste computing resources. In order to prevent redundant calculations in areas without anomalies, the initial screening trigger module extracts single-dimensional features with lower computational requirements for computational truncation. Vegetation cover is a two-dimensional plane parameter that intuitively reflects the macroscopic morphology of surface vegetation. Compared with LAI, which represents three-dimensional structure, and TPP, which represents physiological photosynthesis, vegetation cover responds most directly and rapidly to macroscopic surface damage such as large-scale deforestation and natural fires. At the same time, extracting only vegetation cover does not require calling the complex radiative transfer inversion process. Within the scope of the full-domain pixel matrix, the initial screening trigger module independently extracts the normalized relative density value of vegetation cover for each spatial pixel in the current monitoring phase, and stops the model inversion operation of LAI and TPP for the full-domain pixel matrix.

[0090] The initial screening trigger module reads historical vegetation cover normalized relative density data from multiple periods within a historical time series smoothing window. To eliminate the baseline bias caused by random climate fluctuations in a single year, the initial screening trigger module calculates the historical average value of vegetation cover characteristics within the historical time series smoothing window. The formula for calculating the historical average value of vegetation cover characteristics is as follows:

[0091] ;

[0092] in, Represents spatial coordinate position quantity Historical average of vegetation cover characteristics; This indicates the total number of observation phases included within the historical time series smoothing window. The total number of historical observation phases is usually selected from the number of cloudless remote sensing images of the same phenological period within 3-5 years prior to the target monitoring phase. Indicates the historical observation phase; Indicates the first A historical observation phase; Represents spatial coordinate position quantity In the historical observation phase The extracted vegetation cover normalized relative density values.

[0093] The initial screening trigger module calculates the one-dimensional negative deviation of the current vegetation cover of spatial pixels from the historical average vegetation cover. The one-dimensional negative deviation is used to characterize the degree of degradation of the current apparent vegetation cover relative to the historical normal state. Since the degradation of vegetation parameters is manifested as a decrease in observed values, the smaller the calculated one-dimensional negative deviation, the greater the potential ecological risk reflected. The formula for calculating the one-dimensional negative deviation is as follows:

[0094] ;

[0095] in, Represents spatial coordinate position quantity The one-dimensional negative deviation; Indicates the current monitoring time phase; Represents spatial coordinate position quantity In the current monitoring phase The obtained normalized relative density values ​​of vegetation cover.

[0096] The initial screening trigger module sets a first-order cutoff lower limit threshold. This threshold serves as a preliminary boundary to distinguish between the normal physiological fluctuation range of the system and the potential risk of degenerative abnormalities. To maintain a high initial screening tolerance and prevent the underreporting of genuine minor anomalies, the constant value of the first-order cutoff lower limit threshold is set between 0.10 and 0.20. In the specific value setting operation, the specific value of the first-order cutoff lower limit threshold is adaptively set based on the statistical standard deviation of the natural fluctuation state of vegetation in the target monitoring area over historical periods. Typically, one times the standard deviation of historical natural fluctuations is taken as the first-order cutoff lower limit threshold to ensure that the vast majority of normal phenological fluctuations are effectively filtered out. The initial screening trigger module compares the one-dimensional negative deviation with the first-order cutoff lower limit threshold to construct a trigger discriminator function. That is, it is triggered when the negative value of the one-dimensional negative deviation exceeds the preset lower limit threshold range. The mathematical expression of the trigger discriminator function is as follows:

[0097] ;

[0098] in, Represents spatial coordinate position quantity The corresponding output value of the trigger discriminator function; This represents the set lower threshold for first-order truncation, and Limited to positive real numbers; Represents logical operations; Represents spatial coordinate position quantity The one-dimensional negative deviation.

[0099] The initial screening trigger module executes the classification routing distribution logic based on the output value of the trigger discriminator function. When the output value of the trigger discriminator function is 0, it indicates that the single-dimensional negative deviation has not reached the first-order truncation lower limit threshold, and determines that the spatial pixel is currently in the normal fluctuation range. The initial screening trigger module blocks the calculation path of the spatial pixel with the output value of the trigger discriminator function being zero to the high-dimensional inspection module, marks the corresponding spatial pixel as a normal background pixel, and records the spatial coordinate system attributes.

[0100] When the output value of the trigger discriminator function is one, it indicates that the negative deviation of the single dimension has reached or exceeded the first-order truncation lower limit threshold, and it is determined that the spatial pixel has a potential risk of vegetation degradation. The initial screening trigger module collects the spatial coordinate system attributes of the spatial pixels whose output value of the trigger discriminator function is one, and aggregates all spatial pixels belonging to this class to generate a first-level suspected set. The initial screening trigger module sends the spatial pixel coordinate index contained in the first-level suspected set to the high-dimensional verification module, thereby completing the adaptive scheduling of computing power and spatial dimensionality reduction operation.

[0101] See attached document Figure 2 In this embodiment of the invention, the specific steps of the high-dimensional verification module in performing suspected region computing power scheduling, high-dimensional parameter joint inversion, historical steady-state variation coefficient extraction, and asymmetric weight matrix design are as follows:

[0102] The high-dimensional verification module receives the first-level suspected set sent by the initial screening trigger module. The first-level suspected set contains the spatial coordinate system attributes of spatial pixels that show initial signs of degradation. The single-dimensional vegetation cover parameter can only characterize the spread of vegetation in the two-dimensional horizontal direction, and cannot fully reflect the damage to the vertical structure inside the vegetation canopy and the potential decline in the physiological function of vegetation photosynthesis and carbon sequestration. In order to determine the authenticity of ecological degradation, the high-dimensional verification module needs to introduce the leaf area index parameter, which characterizes the canopy structure, and the total primary productivity parameter, which characterizes the physiological function, for multi-dimensional joint verification.

[0103] S320, the high-dimensional verification module, based on the spatial coordinate system attributes of the spatial pixels recorded in the first-level suspected set, calls the radiative transfer model and ecological process model within the set suspected spatial coordinate constraints. The high-dimensional verification module only inverts the current leaf area index parameter and the current total primary productivity parameter for the spatial pixels in the first-level suspected set. The high-dimensional verification module performs the same relative density calculation and extreme value normalization processing on the current leaf area index parameter and the current total primary productivity parameter to obtain the current leaf area index normalized relative density value and the current total primary productivity normalized relative density value of the corresponding spatial pixels. The high-dimensional verification module linearly combines the current vegetation cover normalized relative density value, the current leaf area index normalized relative density value, and the current total primary productivity normalized relative density value to generate the current three-dimensional normalized feature vector of the current monitoring phase.

[0104] S330, the high-dimensional verification module extracts the three-dimensional normalized feature vector time series of spatial pixels within the historical observation period. Before statistical calculation, the high-dimensional verification module reads the quality assessment mask file attached to the historical remote sensing images, removing invalid observation phase data that is obscured by clouds, covered by cloud shadows, or contaminated by strong atmospheric aerosols, ensuring that the historical observation series used in the calculation represents the true physiological state of vegetation. The high-dimensional verification module calculates the historical mathematical mean and historical standard deviation of each dimension of ecological parameter within the time series after invalid data removal. The high-dimensional verification module aggregates the historical mathematical mean of each dimension of ecological parameter to generate a three-dimensional mean vector, and aggregates the historical standard deviation of each dimension of ecological parameter to generate a three-dimensional standard deviation vector. The high-dimensional verification module calculates the ratio of the historical standard deviation of a specific dimension parameter within the three-dimensional standard deviation vector to the historical mathematical mean of the corresponding dimension parameter within the three-dimensional mean vector, obtaining the historical steady-state coefficient of variation of the corresponding dimension of ecological parameter. The formula for calculating the historical steady-state coefficient of variation is as follows:

[0105] ;

[0106] in, Represents spatial coordinate position quantity In the Historical steady-state variation coefficients in each dimension; Represents spatial coordinate position quantity In the Historical standard deviation of each dimension; Represents spatial coordinate position quantity In the Historical mathematical mean of each dimension; Dimensions representing ecological parameters.

[0107] S340, the high-dimensional verification module constructs an asymmetric weighting factor based on the historical steady-state coefficient of variation. The historical fluctuation amplitude of ecological parameters in different dimensions varies significantly under the natural environmental background. For example, the total primary productivity parameter is more severely affected by interannual meteorological precipitation and temperature fluctuations, exhibiting large natural periodic noise. The vegetation cover parameter shows high numerical stability under undisturbed conditions. By introducing the reciprocal of the coefficient of variation as the weight of the spatial distance metric, it is essentially to perform non-equidistant stretching of the multidimensional feature space. During the calculation process, parameters with large natural fluctuations are given smaller weights to prevent normal meteorological fluctuations from being misjudged as abnormal. At the same time, parameters with stable historical performance are given larger weights, so that once a stable parameter experiences a slight decrease, its penalty value in the distance metric will increase exponentially, thereby improving the algorithm's sensitivity to the real ecological degradation during the incubation period.

[0108] To prevent system collapse caused by the extreme stability of individual ecological parameters leading to historical standard deviations approaching zero, which in turn results in infinitely large reciprocal calculations, the high-dimensional testing module sets a minimum positive real constant as a cutoff limit before calculating the reciprocal of the coefficient of variation. When the extracted historical steady-state coefficient of variation is less than this minimum positive real constant, the historical steady-state coefficient of variation is forcibly replaced by the minimum positive real constant. The formula for calculating the asymmetric weighting factor is as follows:

[0109] ;

[0110] in, Represents spatial coordinate position quantity In the Asymmetric weighting factors in each dimension; Represents spatial coordinate position quantity In the The reciprocal of the historical steady-state coefficient of variation in each dimension; Indicates the dimension of the summation operation; Represents spatial coordinate position quantity In the The reciprocal of the historical steady-state coefficient of variation in each dimension.

[0111] S350, the high-dimensional verification module, according to the arrangement order of the ecological parameter dimension identifiers, places the asymmetric weight factors calculated for each of the three dimensions on the main diagonal, sets the elements in the off-diagonal positions to zero, and constructs a third-order square matrix of asymmetric diagonal weights. The formula for constructing the asymmetric diagonal weight matrix is ​​as follows:

[0112] ;

[0113] in, Represents spatial coordinate position quantity The corresponding asymmetric diagonal weight matrix; This indicates the symbol for constructing diagonal matrices; Asymmetric weighting factors representing the vegetation cover dimension; This represents the asymmetric weighting factor for the leaf area index dimension; This represents the asymmetric weighting factor for the total primary productivity dimension.

[0114] The high-dimensional verification module stores the generated asymmetric diagonal weight matrix in the system memory to support the subsequent weighted Mahalanobis distance calculation process under the multicollinearity defense mechanism.

[0115] See attached document Figure 2 In this embodiment of the invention, the specific steps of the high-dimensional testing module in performing regularized covariance calculation, weighted Mahalanobis feature distance measurement, and anomaly detection under multicollinearity defense are as follows:

[0116] The high-dimensionality test module calculates the third-order historical covariance matrix of a single spatial pixel based on the three-dimensional normalized feature vector time series within the historical observation period. In vegetation ecosystems, the three ecological parameters—vegetation cover, leaf area index, and total primary productivity—exhibit a high degree of positive correlation in biological mechanisms. This high degree of multicollinearity can cause the third-order historical covariance matrix to approach a singular matrix, easily leading to matrix inversion failure or divergence in subsequent Mahalanobis distance calculations. To prevent computational system crashes caused by multicollinearity, the high-dimensionality test module adds a minimal positive real number regularization parameter to the main diagonal elements of the third-order historical covariance matrix, generating a regularized covariance matrix. The value of the minimal positive real number regularization parameter is set to 10. -6 The high-dimensional testing module can enforce the full-rank invertibility property of a matrix without changing the original data space distribution structure. The formula for calculating the regularized covariance matrix is ​​as follows:

[0117] ;

[0118] in, Represents spatial coordinate position quantity The regularized covariance matrix; Represents spatial coordinate position quantity The third-order historical covariance matrix is ​​generated based on historical observation time series; This represents the set regularization parameter for the smallest positive real number; This represents a third-order identity matrix.

[0119] The high-dimensional verification module reads the current-period three-dimensional normalized feature vector, three-dimensional mean vector, asymmetric diagonal weight matrix, and the inverse of the regularized covariance matrix from the current monitoring phase. It then embeds the asymmetric diagonal weight matrix into the standard Mahalanobis distance measurement space to calculate the weighted Mahalanobis distance for a specific spatial pixel. Introducing the asymmetric diagonal weight matrix into the standard Mahalanobis distance formula is geometrically equivalent to scaling the original three-dimensional feature space non-uniformly across the spatial coordinate axes. By multiplying the weight matrix on both sides of the distance calculation, the system actively amplifies the contribution of historically stable ecological parameter dimensions to the distance calculation while suppressing numerical interference from dimensions with high natural noise. This approach comprehensively considers the differences in the natural fluctuation amplitudes of different ecological parameter dimensions and the coupling correlation between multidimensional parameters, outputting a comprehensive spatial distance scalar value representing the deviation of the current vegetation state from the historical steady-state center. The formula for calculating the weighted Mahalanobis distance is as follows:

[0120] ;

[0121] in, Represents spatial coordinate position quantity The weighted squared Mahalanobis distance term value; Represents spatial coordinate position quantity In the current monitoring phase The current period three-dimensional normalized feature vector; Represents spatial coordinate position quantity The three-dimensional mean vector; This represents the matrix transpose operator. Represents spatial coordinate position quantity The corresponding asymmetric diagonal weight matrix; Represents the regularized covariance matrix The inverse matrix is ​​generated after matrix inversion.

[0122] The high-dimensional test module sets the critical threshold for the chi-square distribution distance. Since the three-dimensional normalized feature vector contains three independent basic ecological dimension parameters, the squared term of the weighted Mahalanobis distance approximately follows a chi-square distribution with three degrees of freedom in mathematical statistics. Based on the preset confidence level, the high-dimensional test module determines the corresponding critical threshold for the chi-square distribution distance by consulting the standard chi-square distribution table. In conventional ecological monitoring applications, the confidence level is usually set to 95% to balance the sensitivity of anomaly identification and the need for false alarm suppression. The specific value of the corresponding critical threshold constant for the chi-square distribution distance is 7.815.

[0123] The high-dimensional inspection module extracts the weighted Mahalanobis distance squared term value and performs a numerical comparison logic with the chi-square distribution distance critical threshold. If the weighted Mahalanobis distance squared term value is less than or equal to the chi-square distribution distance critical threshold, the high-dimensional inspection module determines that the comprehensive ecological fluctuation state of the spatial pixel is still within the multidimensional historical confidence ellipsoid, blocks the calculation path of the spatial pixel to the abnormal record node, and classifies the spatial pixel as normal background.

[0124] When the weighted Mahalanobis distance squared term exceeds the critical threshold of the chi-square distribution distance, the high-dimensional inspection module determines that the spatial pixel has undergone a real ecological abnormality. The high-dimensional inspection module filters spatial pixels whose weighted Mahalanobis distance exceeds the critical threshold of the chi-square distribution distance and uniformly classifies the spatial pixels that meet the screening conditions into the secondary candidate anomaly set. The high-dimensional inspection module sends the secondary candidate anomaly set and the weighted Mahalanobis distance data associated with each spatial pixel in the secondary candidate anomaly set to the background stripping module, providing a high-confidence target area set for the subsequent spatial background denoising process.

[0125] See attached document Figure 2 In this embodiment of the invention, the background stripping module performs a spatial homogeneous meteorological background stripping operation with accompanying disaster degradation routing. Specifically, the steps of spatial homogeneous background control point constraint retrieval and branch routing logic control mechanism are as follows:

[0126] The secondary candidate anomaly set output by the high-dimensional inspection module may contain false degradation signals caused by large-scale regional meteorological and environmental fluctuations. The decline in vegetation parameters caused by large-scale regional meteorological and environmental fluctuations is spatially universal and homogeneous. The background stripping module performs spatial homogeneous meteorological background stripping operation on specific spatial pixels within the secondary candidate anomaly set. The background stripping module reads the spatial coordinate attributes of specific spatial pixels in the secondary candidate anomaly set and establishes a spatial neighborhood retrieval window with the spatial coordinates of the specific spatial pixel as the geometric center. The physical size of the spatial neighborhood retrieval window is set according to the actual scale requirements of ecological monitoring operations and the homogeneity scale of regional climate. It is usually set as a rectangular spatial range with a side length spanning three to five kilometers. In the pixel coordinate system, the background stripping module reads the spatial resolution value of the input remote sensing image, divides the physical size of the spatial neighborhood retrieval window by the spatial resolution value, and calculates the number of pixel rows and columns covered by the spatial neighborhood retrieval window, thereby constructing a rectangular pixel retrieval array centered on the specific spatial pixel in memory.

[0127] The background stripping module performs spatial homogeneous baseline control point constraint retrieval. It traverses all neighboring spatial pixels within the spatial neighborhood retrieval window and applies multiple conditions for homogeneous filtering. These conditions require neighboring spatial pixels to simultaneously meet three parallel attribute constraints: First, the ecological functional zoning index assigned to a neighboring spatial pixel must be completely consistent with the ecological functional zoning index of the specific spatial pixel; second, the vegetation type index assigned to a neighboring spatial pixel must be completely consistent with the vegetation type index of the specific spatial pixel; third, the neighboring spatial pixel must belong to the normal background pixels that were blocked and discarded by the initial screening trigger module, meaning the neighboring spatial pixel has not been classified into the first-level suspected set. The background stripping module sets neighboring spatial pixels that simultaneously meet the above three constraints as baseline control points. These baseline control points represent normal vegetation reference samples that have not experienced abnormal fluctuations under the same geographical environment and vegetation category.

[0128] The background stripping module counts the total number of baseline control points identified within the spatial neighborhood retrieval window. During this process, the calculation system iterates through the pixels in the retrieval array one by one using an indicator function. Pixels that simultaneously meet the conditions of region consistency, category consistency, and normal status are judged using Boolean criteria, and the number of pixels meeting the conditions is accumulated. The mathematical model for extracting the total number of baseline control points is as follows:

[0129] ;

[0130] in, Represents spatial coordinate position quantity Total number of baseline control points extracted centered on the target area; Represents the position quantity around spatial coordinates The set of discrete pixels contained within the constructed spatial neighborhood retrieval window; Indicates the spatial coordinate position of the neighboring area; The ecological functional zoning consistency indicator function is expressed in terms of the spatial coordinates of the adjacent areas. Spatial coordinate position quantity Output 1 if they belong to the same category of ecological function zones, and output 0 if they do not belong to the same category of ecological function zones. This represents a vegetation type consistency indicator function, measured in neighboring spatial coordinates. Spatial coordinate position quantity Output 1 if they belong to the same vegetation type, and output 0 if they do not belong to the same vegetation type. This represents a normal background indicator function, measured at nearby spatial coordinates. Output 1 if the child is not classified into the first-level suspected set, and output 0 if the child is classified into the first-level suspected set.

[0131] The background stripping module introduces a preset effective number control line. This preset effective number control line is used to determine whether there is a baseline reference sample within the spatial neighborhood search window that supports statistical operations. To ensure the reliability of the subsequent calculation of the mean of background fluctuations and to conform to the large sample center limit theorem, the preset effective number control line is usually set to 30-50 pixels. The background stripping module extracts the total number of baseline control points and compares the total number of baseline control points with the preset effective number control line. This comparison serves as the trigger condition for executing the branch routing logic control mechanism.

[0132] When the total number of identified baseline control points reaches or exceeds the preset effective retrieval control line, the background stripping module determines that there is sufficient undisturbed homologous normal vegetation around a specific spatial pixel. Sufficient homologous normal vegetation indicates that the ecological baseline of the current area has not experienced a systemic collapse, providing a basis for comparison to implement refined meteorological background stripping. The background stripping module imports the specific spatial pixel and all baseline control point data associated with the specific spatial pixel into the meteorological background stripping routine calculation route, triggering subsequent statistical and deviation stripping nodes for the characteristic distance of baseline control points.

[0133] If the total number of identified baseline control points does not reach the preset effective retrieval control line, the background stripping module determines that there is a severe lack of normal homologous reference vegetation around a specific spatial pixel. The physical cause of the lack of homologous normal vegetation is usually that the area has encountered extreme weather disasters or has experienced large-scale man-made destructive damage, resulting in almost all vegetation in the spatial neighborhood retrieval window being in an abnormal degradation state. In the scenario of large-scale damage across the entire area, if the very few remaining pixels are forcibly extracted as baseline references for background stripping comparison, the abnormal monitoring signal will be unreasonably smoothed out and canceled, resulting in anomaly underreporting. In response to this special case of lack of reference benchmarks, the background stripping module triggers a disaster degradation route, i.e., the direct anomaly determination path.

[0134] As a fault-tolerant protection channel for extreme ecological disaster scenarios, the disaster degradation routing maintains the sensitivity of anomaly warning by blocking the subsequent benchmark subtraction logic. When entering the disaster degradation routing state, the background stripping module stops the calculation process of the average distance of regional background fluctuations for specific spatial pixels, forcibly skips the net anomaly deviation calculation step, and directly identifies the high-dimensional state deviation of specific spatial pixels as real disaster degradation anomalies, and directly loads the specific spatial pixels into the confirmed anomaly pixel set. By configuring this branch routing logic control mechanism, the background stripping module can eliminate meteorological false fluctuations while taking into account the effective detection rate under extreme disaster conditions across the entire domain.

[0135] See attached document Figure 2In this embodiment of the invention, the specific steps for the background stripping module to calculate the net anomaly deviation under local anomaly verification are as follows:

[0136] After importing the data of a specific spatial pixel and all associated background control points into the conventional meteorological background stripping calculation route, the background stripping module extracts the weighted Mahalanobis distance squared term of all associated background control points. The module then calculates the arithmetic mean of these weighted Mahalanobis distance squared terms to generate the regional background fluctuation mean distance. This regional background fluctuation mean distance reflects the general baseline shift caused by normal vegetation under the influence of large-scale meteorological factors in the current monitoring phase. By extracting the regional background fluctuation mean distance, the background stripping module constructs a common reference datum for stripping false signals. The formula for calculating the regional background fluctuation mean distance is as follows:

[0137] ;

[0138] in, Represents spatial coordinate position quantity The average distance of the associated regional background fluctuations; Represents spatial coordinate position quantity Total number of baseline control points extracted centered on the target area; Represents the position quantity around spatial coordinates The constructed set of baseline control points; Indicate the baseline control points; This indicates traversing the baseline control points. The weighted squared Mahalanobis distance term value.

[0139] The background stripping module calculates the net anomaly deviation of a specific spatial pixel. It subtracts the calculated average distance of regional background fluctuations from the weighted Mahalanobis distance squared term of the pixel. In real natural spatial distribution, if the weighted Mahalanobis distance squared term of a specific spatial pixel is less than the average distance of regional background fluctuations, the direct subtraction will result in a negative real number. The appearance of a negative real number indicates that the degree of anomaly fluctuation of the specific spatial pixel is lower than the average fluctuation level of the surrounding normal vegetation, and that the specific spatial pixel does not have a risk of degradation. To unify the data storage format and avoid interference from negative numbers in subsequent judgment logic, the background stripping module nests a maximum value truncation function outside the subtraction formula, setting the subtraction result less than 0 to 0. The formula for calculating the net anomaly deviation is as follows:

[0140] ;

[0141] in, Represents spatial coordinate position quantity Net abnormal deviation; This indicates the operator for truncating the maximum value in a function; Represents spatial coordinate position quantity The weighted squared Mahalanobis distance term value at the current monitoring time phase; Represents spatial coordinate position quantity The average distance of the associated regional background fluctuations; It represents the constant zero.

[0142] The background stripping module sets a net anomaly control threshold and performs local anomaly verification based on the net anomaly deviation. The net anomaly control threshold is used to measure whether the remaining independent net anomaly distance after completely stripping away the influence of general meteorological background reaches the degradation level that requires an alarm. Since the net anomaly deviation has removed most of the natural noise base, the value of the net anomaly control threshold must be lower than the chi-square distribution distance critical threshold used in the high-dimensional verification stage. The empirical range of the net anomaly control threshold is set between 3.0 and 5.0. In actual engineering deployment, the background stripping module selects a specific value based on the sensitivity of the ecological functional zoning of the specific spatial pixel. For highly sensitive areas such as the core area of ​​national nature reserves, the net anomaly control threshold is set to the lower limit of 3.0 to improve the anomaly detection rate; for general ecological red line periphery areas, the net anomaly control threshold is set to the upper limit of 5.0 to control false alarms.

[0143] The background stripping module compares the net anomaly deviation with the net anomaly control threshold. If the net anomaly deviation is less than or equal to the net anomaly control threshold, the background stripping module determines that the high-dimensional spatial anomaly signal exhibited by a specific spatial pixel is mainly due to pseudo-degradation caused by regional meteorological fluctuations, and removes the specific spatial pixel from the secondary candidate anomaly set. If the net anomaly deviation is greater than the net anomaly control threshold, the background stripping module determines that the specific spatial pixel has significant local ecological damage that deviates from the meteorological benchmark, and loads the specific spatial pixel into the confirmed anomaly pixel set.

[0144] To filter out isolated pixel noise caused by random hardware failures of remote sensing sensors, the background stripping module performs spatial connectivity calculation on the confirmed abnormal pixel set. By judging the adjacency relationship of pixels in spatial coordinates, the background stripping module calculates the number of pixels that constitute a connected region. The background stripping module sets a minimum abnormal contiguous area threshold, which is usually set to four consecutive pixels. The background stripping module deletes isolated pixels with fewer connected pixels than the minimum abnormal contiguous area threshold from the confirmed abnormal pixel set, completing the final purification of the warning data. The purified confirmed abnormal pixel set will be output to the front-end display system to generate the final ecological violation warning report.

[0145] See attached document Figure 2In this embodiment of the invention, the specific steps of the map generation module in performing spatial morphological correction and outputting the spatiotemporal evolution results of ecological anomalies are as follows:

[0146] The map generation module receives the confirmed anomaly pixel set sent by the background stripping module. Since the confirmed anomaly pixel set is obtained based on multi-dimensional feature distance measurement and background stripping calculation at the independent pixel level, the confirmed anomaly pixel set is prone to exhibiting jagged edge features in spatial distribution. Furthermore, there may be tiny holes left inside the confirmed anomaly pixel set due to individual pixels not reaching the net anomaly control threshold. In order to generate smooth patch boundaries that conform to the geographic information system specifications, the map generation module performs spatial morphological correction on the confirmed anomaly pixel set. The map generation module establishes a two-dimensional binary image matrix that is completely aligned with the spatial coordinate system and spatial resolution of the original input remote sensing image. The map generation module assigns a value of 1 to the corresponding position of the spatial pixel contained in the confirmed anomaly pixel set in the two-dimensional binary image matrix, and assigns a value of 0 to the remaining positions in the two-dimensional binary image matrix.

[0147] The image generation module performs mathematical morphological closing operations on the two-dimensional binary image matrix. This closing operation is a cascaded combination of dilation and erosion operations. The module presets morphological structural elements, the size of which is set based on the spatial resolution of the remote sensing image and the minimum physical area to be filled required for ecological monitoring. For standard 30-meter resolution images, the morphological structural elements are set as a 3×3 pixel cross-shaped matrix. For higher resolution images, the side length and number of pixels of the morphological structural elements are increased proportionally to ensure that the actual physical area processed by the system remains constant when eliminating holes. The image generation module uses the morphological structural elements to first perform dilation operations on the two-dimensional binary image matrix, followed by erosion operations. In morphological processing, dilation expands the boundaries of anomalous pixels outward, filling tiny holes inside anomalous patches and connecting broken adjacent pixels. Erosion shrinks the boundaries inward using the same morphological structural elements, restoring the outer contour of the anomalous patches to their original size while maintaining the filled state, and eliminating jagged edges. The calculation formula for mathematical morphological closing is as follows:

[0148] ;

[0149] in, This represents the corrected two-dimensional binary image matrix output after the closing operation. This represents the original two-dimensional binary image matrix input to the map generation module; Represents the defined morphological structural elements; Symbols representing mathematical morphological dilation operations; This represents the symbol for mathematical morphological erosion operations.

[0150] The image generation module performs raster vectorization on the corrected two-dimensional binary image matrix. The image generation module converts the corrected two-dimensional binary image matrix into a current period anomalous patch vector polygon containing ordered spatial coordinate nodes. The current period anomalous patch vector polygon has a clear geometric topological relationship and can be directly read and stored by conventional spatial databases.

[0151] The map generation module performs spatiotemporal evolution feature calculations for ecological anomalies. Connecting to a historical spatial database, it extracts vector polygons of historical anomaly patches stored from the previous monitoring phase. Under a unified spatial reference coordinate system, the module performs polygon topological overlay calculations on the vector polygons of current and historical anomaly patches. By extracting spatial differences and intersections, the module calculates the areas of newly added and persistent anomaly regions, representing the dynamics of spatial evolution. The formula for calculating the area of ​​newly added anomaly regions is as follows:

[0152] ;

[0153] The formula for calculating the area of ​​a persistent anomaly region is as follows:

[0154] ;

[0155] in, This represents the area value of the newly added anomaly region obtained through spatial overlay calculation; This represents the area of ​​the persistent anomaly region obtained through spatial overlay calculation; This represents the set of two-dimensional closed regions covered by the vector polygons of current anomaly patches in a Cartesian coordinate system. This represents the set of two-dimensional closed regions covered by the vector polygons of historical anomalous patches in a Cartesian coordinate system. This represents the set difference operation symbol, used to extract the spatial portion of anomaly vector polygons that belong to the current period but not to the historical period. This symbol represents the intersection operation, used to extract the spatial overlap between vector polygons of current period anomalies and vector polygons of historical period anomalies. and Let represent the differential area elements along the horizontal and vertical directions in the Cartesian coordinate system, respectively. This symbol represents a two-dimensional area double integral calculation performed within a specified spatial operation area.

[0156] The map generation module integrates the spatial location information of the vector polygons of current abnormal patches, the area of ​​newly added abnormal areas, the area of ​​persistent abnormal areas, and the maximum value of the associated net anomaly deviation to generate a structured data format of the spatiotemporal evolution results of ecological anomalies. The map generation module packages the spatiotemporal evolution results of ecological anomalies into an ecological violation early warning report. The map generation module pushes the ecological violation early warning report to the front-end display system through a network communication interface. After receiving the ecological violation early warning report, the front-end display system uses a geographic information rendering engine to highlight and overlay the vector polygons of the current abnormal patches on the electronic map base map, thereby realizing the location and visual warning of ecological degradation events.

[0157] Specific application examples:

[0158] To verify the effectiveness and advancement of the rapid monitoring method for vegetation anomalies within ecological protection red lines proposed in this invention in practical and complex engineering projects, this embodiment takes a suspected vegetation degradation event in a typical ecological protection red line area as an example, strictly following the order of the attached figures ( Figures 3 to 5 The specific implementation process and beneficial effects of the present invention will be explained in detail by combining specific mathematical formula derivation and actual numerical substitution.

[0159] Comparison test of the optimization efficiency and resource consumption of multiple algorithms (combined) Figure 3 ):

[0160] To objectively evaluate the overall performance of the algorithm of this invention, this embodiment conducted long-term monitoring and comparison of the entire spatial domain, such as... Figure 3 As shown, traditional single-dimensional monitoring methods consume significant computing resources (normalized to 100.0%), and the false alarm rate during drought periods is as high as 35.4%. To address this issue, this invention introduces a computing power blocking mechanism in the initial screening trigger module.

[0161] For the undamaged background pixels that occupy the majority of the image area, the system calculates the deviation using a one-dimensional negative deviation formula:

[0162] ;

[0163] In this example, the current monitored vegetation cover of a typical normal pixel. The average vegetation cover in its historical time series Substituting into the formula, we obtain the one-dimensional negative deviation:

[0164] ;

[0165] The system sets a first-order truncation lower threshold. And substitute it into the trigger discriminator function:

[0166] ;

[0167] because The output of the discriminator function is triggered. Based on this, the system determines that the pixel is normal, directly blocking the calculation path from that pixel to the high-dimensional inversion model. Because a large number of normal pixels are effectively filtered out in the initial screening stage, the computational resource consumption of this invention is reduced to only 12.0% (e.g., Figure 3 As shown in the figure, thanks to subsequent high-dimensional verification, its anomaly detection rate rose to 96.2%, and the false alarm rate during drought periods dropped to 4.1%, achieving a performance breakthrough.

[0168] Temporal ecological parameter normalization feature extraction process (combined with) Figure 4 ):

[0169] For target pixels initially screened as potentially degraded, the high-dimensional inspection module performs... Three-dimensional ecological parameters were extracted using a temporal orientation method and then normalized to achieve the desired results. Figure 4 Mid-dimensional Taking (total primary productivity) as an example, first calculate the spatial parameter relative density: ;

[0170] Extract the absolute value of the current raw observations The historical baseline maximum value corresponding to the ecological type Substituting into the formula, we get:

[0171] ;

[0172] Then, the extreme value normalization formula was applied to find the historical minimum relative density of the area. The highest relative density in history : ;

[0173] Substitute numerical values ​​and solve: ;

[0174] Similarly, extract and substitute into the calculation to obtain the dimension. (Vegetation coverage) and dimensions (Leaf Area Index) Substitute this into the formula for constructing three-dimensional normalized feature vectors: ;

[0175] This generates Figure 4 The specific eigenvector column matrix of the current time phase: ;

[0176] like Figure 4 As shown at the end of the broken line, this multidimensional vector accurately unifies the dimensions, presenting a coordinated decrease in the core ecological parameters at the current time phase.

[0177] Mahalanobis distance measurement and net anomaly deviation calculation under local anomaly verification (combined with) Figure 5 )

[0178] To confirm whether the abrupt drop constitutes genuine degradation, the system introduces a coefficient of variation to assign spatial weights. Figure 4 Extremely stable vegetation cover ( For example, its historical mathematical mean Historical standard deviation Substituting into the historical steady-state coefficient of variation formula: ;

[0179] get A smaller coefficient of variation assigns a matrix weight to this dimension. Combining the asymmetric diagonal weight matrix and the regularized covariance matrix, the above data is substituted into the weighted Mahalanobis distance formula for multidimensional measurement:

[0180] ;

[0181] Due to the exponential penalty caused by the sharp decline in stability parameters, the system calculates the current weighted Mahalanobis distance as follows: .like Figure 5 As shown, this value far exceeds the chi-square distribution critical threshold (7.815), triggering an anomaly alarm.

[0182] To remove background noise caused by widespread abnormal weather (such as drought periods), a background stripping operation is performed. The system retrieves noise within the spatial neighborhood window. A healthy baseline control point is identified, and the average distance of the regional background fluctuation is calculated by substituting these points into the formula. ;

[0183] The total distance to the surrounding 40 normal reference pixels is calculated to be 140.0. Substituting this into the equation, we get:

[0184] ;

[0185] This indicates that large-scale meteorological fluctuations only resulted in a uniform shift of 3.50. Finally, substituting into the net anomaly deviation formula completely eliminated false meteorological signals:

[0186] ;

[0187] Substitute specific values:

[0188] ;

[0189] like Figure 5As shown by the vertical net anomaly deviation line segment marked in the figure, the net deviation of 41.44 after denoising is still greater than the net anomaly control threshold, thus eliminating climate interference and confirming that the spatial pixel has experienced real surface damage. As can be seen from the specific numerical derivation, this invention solves the problem of high false alarm rate of traditional algorithms.

Claims

1. A method for rapid monitoring of vegetation anomalies along ecological protection red lines, characterized in that, Includes the following steps: Acquire optical remote sensing image data and process it to form a pixel matrix composed of multiple spatial pixels; For each spatial pixel in the pixel matrix, the normalized relative density of vegetation cover is calculated as a one-dimensional feature, and the one-dimensional negative deviation is calculated based on the one-dimensional feature. The one-dimensional negative deviation is compared with the preset first-order truncation lower limit threshold to filter and generate a first-level suspected set. For the pixels in the first-level suspected set, calculate the normalized relative density of leaf area index and the normalized relative density of total primary productivity, and construct a three-dimensional normalized feature vector together with the normalized relative density of vegetation cover. For the pixels in the first-level suspected set, based on the three-dimensional normalized feature vector, an asymmetric diagonal weight matrix is ​​constructed according to the historical steady-state variation coefficient and the weighted Mahalanobis distance squared term is solved to filter pixels that exceed the preset chi-square distribution distance critical threshold and generate a second-level candidate anomaly set. Within the spatial neighborhood of a pixel in the secondary candidate anomaly set, homologous normal background pixels are searched as background control points, and the total number of background control points is counted. Based on whether the total number of the baseline control points reaches the preset effective number control line, the branch routing logic is executed to generate a set of confirmed abnormal pixels. Spatial morphological correction and raster vectorization transformation are performed on the confirmed abnormal pixel set to output a map of the area of ​​abnormal vegetation change along the ecological protection red line.

2. The method for rapid monitoring of vegetation anomalies along ecological protection red lines according to claim 1, characterized in that, The construction of the three-dimensional normalized feature vector of the spatial pixel specifically includes: Based on ecological function zoning data and land cover classification raster data, each spatial cell in the spatial cell matrix is ​​assigned a spatial assessment unit attribute label containing ecological function zoning index and vegetation type index. For each spatial pixel, the evaluation benchmark reference values ​​for vegetation cover, leaf area index, and total primary productivity are obtained. The evaluation benchmark reference values ​​are obtained using an open mechanism: if there are publicly available benchmark value results for the area where the target pixel is located, the publicly available benchmark value results are directly extracted as the evaluation benchmark reference values; if there are no publicly available benchmark value results, then, with the target spatial pixel as the center, neighboring pixels of the same spatial evaluation unit attribute label within a certain spatial range are selected, and the historical maximum values ​​of vegetation cover, leaf area index, and total primary productivity of the neighboring pixels within the historical observation period are extracted, and their average values ​​are calculated and set as the evaluation benchmark reference values. The relative density of spatial parameters is obtained by calculating the division ratio of the observed vegetation cover, leaf area index, and total primary productivity parameters of the current observation phase corresponding to the optical remote sensing image data with the corresponding evaluation benchmark reference values. The extreme value normalization mathematical method is used to transform the relative density of the spatial parameters into a numerical distribution range of zero to one, thereby generating a normalized relative density value. The normalized relative density of vegetation cover, the normalized relative density of leaf area index, and the normalized relative density of total primary productivity of a single spatial pixel are linearly concatenated to form the three-dimensional normalized feature vector.

3. The method for rapid monitoring of vegetation anomalies along ecological protection red lines according to claim 2, characterized in that, The process of comparing the one-dimensional negative deviation with a preset first-order truncation threshold to generate a first-level suspected set specifically includes: Historical optical remote sensing image data of the same phenological period as the spatial pixel are extracted, and the vegetation cover is normalized to the historical relative density data within the historical time series smoothing window. The historical average value of the historical data is calculated to obtain the historical average value of the vegetation cover feature. Calculate the one-dimensional negative deviation of the normalized relative density of vegetation cover from the historical average value of the vegetation cover feature at the current observation time. A trigger discriminator function is constructed to compare the single-dimensional negative deviation with the first-order truncation lower limit threshold. Based on the output value of the trigger discriminator function, the spatial coordinate system attributes of the spatial pixels are collected and aggregated to generate the first-level suspected set.

4. The method for rapid monitoring of vegetation anomalies along ecological protection red lines according to claim 1, characterized in that, The construction of the asymmetric diagonal weight matrix based on the historical steady-state variation coefficient specifically includes: The historical steady-state variation coefficient of the corresponding ecological parameter is obtained by calculating the ratio of the historical standard deviation to the historical mathematical mean of each ecological parameter in the time series after invalid data removal for spatial pixels. When the extracted historical steady-state variation coefficient is less than the set minimum positive real constant, the historical steady-state variation coefficient is equivalently replaced with the minimum positive real constant. The inverse term of the historical steady-state variation coefficient is calculated to obtain the asymmetric weighting factor; The asymmetric weighting factors are placed on the main diagonal, and the elements on the non-main diagonal are set to zero, thus constructing the asymmetric diagonal weighting matrix in the form of a third-order square matrix.

5. The method for rapid monitoring of vegetation anomalies along ecological protection red lines according to claim 4, characterized in that, Based on the three-dimensional normalized feature vector, an asymmetric diagonal weight matrix is ​​constructed according to the historical steady-state variation coefficient, and the weighted Mahalanobis distance squared term is solved. Pixels exceeding the preset chi-square distribution distance critical threshold are screened to generate a secondary candidate anomaly set, specifically including: Based on the time series of three-dimensional normalized feature vectors within the historical observation period, the third-order historical covariance matrix of a single spatial pixel is calculated. A minimal positive real regularization parameter is added to the main diagonal elements of the third-order history covariance matrix to generate a regularized covariance matrix. Read the current period three-dimensional normalized feature vector, three-dimensional mean vector, the asymmetric diagonal weight matrix, and the inverse matrix of the regularized covariance matrix of the current observation phase, and calculate the weighted Mahalanobis distance squared term value of the spatial pixel. If the value of the weighted Mahalanobis distance squared term exceeds the critical threshold of the chi-square distribution distance, spatial pixels are filtered and uniformly classified into the secondary candidate anomaly set.

6. The method for rapid monitoring of vegetation anomalies along ecological protection red lines according to claim 1, characterized in that, The execution branch routing logic, which generates a set of confirmed abnormal pixels, specifically includes: When the total number of the background control points reaches the preset effective number control line, the weighted Mahalanobis distance squared term of all background control points associated with a specific spatial cell is extracted, and the arithmetic mean is calculated to generate the average distance of the background fluctuation in the region. The net anomaly deviation is obtained by subtracting the average distance of the background fluctuation in the region from the weighted squared Mahalanobis distance of a specific spatial cell. Spatial cells with a net anomaly deviation greater than a preset net anomaly control threshold are loaded into the confirmed anomaly cell set. If the total number of baseline control points does not reach the preset effective number control line, the pixels in the secondary candidate anomaly set will be directly loaded into the confirmed anomaly pixel set.

7. The method for rapid monitoring of vegetation anomalies along ecological protection red lines according to claim 6, characterized in that, When the total number of the baseline control points reaches the preset effective retrieval control line, generating the confirmed abnormal pixel set further includes: After loading spatial pixels with a net anomaly deviation greater than the net anomaly control threshold into the confirmed anomaly pixel set, spatial connectivity calculation is performed on the confirmed anomaly pixel set, and isolated pixels with a number of connected pixels less than the minimum anomaly contiguous area threshold are deleted.

8. The method for rapid monitoring of vegetation anomalies along ecological protection red lines according to claim 1, characterized in that, The spatial morphological correction and raster vectorization transformation of the confirmed anomalous pixel set specifically includes: A two-dimensional binary image matrix is ​​established, and the spatial pixel positions within the confirmed abnormal pixel set are assigned a value of one, while the remaining positions are assigned a value of zero. Preset morphological structural elements are used to perform dilation and erosion operations on the two-dimensional binary image matrix to form a mathematical morphological closing operation, thereby obtaining a modified two-dimensional binary image matrix. The modified two-dimensional binary image matrix is ​​subjected to raster vectorization transformation to form a vector polygon closed boundary.

9. A method for rapid monitoring of vegetation anomalies along ecological protection red lines according to claim 8, characterized in that, The closed boundary of the vector polygon is the vector polygon of the current period's abnormal patches, and the method further includes: Connect to a historical spatial database to extract vector polygons of historical anomalous patches; Perform polygon topological overlay calculation on the current period abnormal patch vector polygon and the historical period abnormal patch vector polygon to calculate the area of ​​newly added abnormal region and the area of ​​continuous abnormal region.

10. A method for rapid monitoring of vegetation anomalies along ecological protection red lines according to claim 9, characterized in that, The output map of areas with abnormal vegetation changes along the ecological protection red line specifically includes: By integrating the spatial location information of the current abnormal patch vector polygon, the area of ​​the newly added abnormal region, the area of ​​the continuous abnormal region, and the associated net abnormality deviation attribute, the spatiotemporal evolution result of ecological abnormality is generated. The spatiotemporal evolution results of the ecological anomalies are packaged into an ecological violation early warning report, and the vector polygons of the current period's abnormal patches are superimposed on the spatial geographic reference base map for output.

Images