Desertification monitoring method and system based on multi-source remote sensing data fusion

Abstract

The invention discloses a desertification monitoring method and system based on multi-source remote sensing data fusion, and the method comprises the steps: carrying out the weighted scoring of multi-source satellite remote sensing data according to the spectrum, cloud coverage and time resolution, and automatically switching to a data source with a higher priority when the score is lower than a threshold value. Thirdly, performing initial segmentation on the image by using a multi-scale segmentation algorithm in combination with spectrum and texture features, optimizing boundaries through a region growing algorithm, and merging regions with small spectrum differences into desertification plaques; and extracting construction land expansion and agricultural activity intensity indexes from the time series data, and establishing a quantitative relation model with desertification plaque change. And adopting a Kriging space-time interpolation algorithm for missing data to generate a continuous curved surface. And finally, fusing a multi-period segmentation result and human factor data, and outputting a desertification boundary dynamic change diagram and a human influence weight distribution diagram. According to the invention, accurate monitoring of the desertification process and quantitative evaluation of human influence are realized, and a scientific basis is provided for desertification control.

Description

Technical Field

[0001] This invention belongs to the field of ecological monitoring, and in particular relates to a desertification monitoring method and system based on multi-source remote sensing data fusion. Background Technology

[0002] Desertification, as one of the most serious environmental problems globally, directly threatens the ecological security of one-third of the world's land area. Accurate monitoring of the occurrence and development of desertification is crucial for ecological protection and sustainable development. Remote sensing technology, with its advantages of large-scale and multi-temporal observation, has become the main technical means for desertification monitoring. However, current remote sensing monitoring methods for desertification have significant shortcomings in terms of data processing efficiency and resource allocation.

[0003] Existing methods generally employ a serial processing mode to handle large-scale remote sensing data. When faced with massive amounts of data generated by multiple satellite platforms, the processing speed is severely lagging, making it difficult to meet real-time monitoring needs. Furthermore, most methods lack intelligent data source management mechanisms, failing to dynamically optimize data combinations based on monitoring requirements and data quality, resulting in low data utilization efficiency.

[0004] In the identification of desertification boundaries, traditional methods often rely on single segmentation algorithms, which lack adaptability to complex surface environments and have limited boundary extraction accuracy. Insufficient parallel processing capabilities for multi-source remote sensing data restrict the timeliness of desertification monitoring. Large-scale remote sensing data requires multiple stages, including preprocessing, feature extraction, and fusion analysis. Traditional sequential processing methods cannot fully utilize computing resources, resulting in excessively long processing cycles. This limitation in processing efficiency further exacerbates the complexity of data source allocation. When faced with multiple satellite data sources, there is a lack of effective evaluation and selection mechanisms, hindering intelligent switching and priority management of data sources. The existence of data source allocation problems directly affects the spatiotemporal continuity of monitoring data, especially during data acquisition intervals. Missing monitoring information is difficult to compensate for through effective interpolation and reconstruction methods, leading to time gaps in desertification process monitoring.

[0005] How to construct an efficient parallel computing architecture to achieve real-time processing of multi-source remote sensing data, and at the same time establish an intelligent data source allocation mechanism to ensure the continuity and integrity of monitoring data, has become a key issue in improving the accuracy and timeliness of desertification monitoring. Summary of the Invention

[0006] To address the aforementioned technical problems, this invention provides a desertification monitoring method and system based on multi-source remote sensing data fusion. Specifically, the desertification monitoring method based on multi-source remote sensing data fusion includes:

[0007] Acquire multi-source satellite remote sensing data, and use a preset data quality assessment algorithm to weight and score the spectral resolution, cloud coverage, and temporal resolution of the multi-source satellite remote sensing data to obtain the scoring results;

[0008] A dynamic allocation mechanism is established based on the scoring results. If the current data score is lower than a preset threshold, it will automatically switch to a backup data source with higher priority.

[0009] A multi-scale segmentation algorithm is used to perform initial segmentation on images from selected data sources, and the first segmentation result is generated by combining spectral features and texture features.

[0010] The first segmentation result is optimized based on the region growing algorithm. If the spectral difference between adjacent regions is less than the preset tolerance, they are merged into the same desertification patch.

[0011] We extract indicators of construction land expansion and agricultural activity intensity from time-series data to establish a quantitative relationship model between human factors and desertification patch changes;

[0012] For data in the missing time period, the Kriging spatiotemporal interpolation algorithm is used to generate a continuous surface by combining the spatial distribution of existing data points and the desertification evolution trend.

[0013] By integrating multi-period segmentation results and anthropogenic factor quantification data, a dynamic change map of desertification boundaries and a weight distribution map of anthropogenic influences are output.

[0014] Preferably, the process of acquiring multi-source satellite remote sensing data and using a preset data quality assessment algorithm to perform weighted scoring on the multi-source satellite remote sensing data to obtain the scoring results includes:

[0015] Acquire spectral, cloud cover, and temporal values ​​from multi-source satellite data, and calculate spectral resolution, cloud coverage, and temporal resolution using a pre-defined evaluation method;

[0016] The spectral resolution, cloud coverage, and temporal resolution are normalized according to the weight values. If the cloud cover value exceeds the preset value, the corresponding data source is removed.

[0017] The normalized indicators are combined using a weighted method to obtain a score value, and the score value and the corresponding multi-source data from the data source are stored.

[0018] Preferably, the process of establishing a dynamic allocation mechanism based on the scoring results, whereby if the current data score is lower than a preset threshold, the process of automatically switching to a higher-priority backup data source includes:

[0019] Obtain real-time performance metrics data from the data source; collect key parameters such as data transmission latency, connection stability, and response time through the monitoring system to obtain comprehensive performance status information of the current data source.

[0020] Based on the comprehensive performance status information, a weighted average algorithm is used to calculate the real-time score of the data source, thereby obtaining a standardized data source quality score.

[0021] Determine if the real-time score is below a preset threshold. If so, automatically trigger the backup resource activation mechanism to obtain available resource information from a pre-configured backup data source list and determine the current status of the backup resources.

[0022] Available backup resources are sorted using a priority ranking algorithm. Priority scores are calculated based on resource capacity, historical stability, and geographical location to obtain a priority sequence of backup resources.

[0023] Select the highest-ranked backup data source as the switching target according to the priority sequence, establish a connection channel with the target data source, and obtain connection status confirmation information.

[0024] If the connection status confirmation information shows that the connection is successful, a data stream switching operation is performed to redirect the business request from the original data source to the target backup data source, thus completing the dynamic switching process of the data source.

[0025] A continuous monitoring mechanism is used to track the performance of the switched data source, record the execution time of the switching operation, data integrity and business continuity, determine the success status of the switching operation and update the system operation log.

[0026] Preferably, the process of using a multi-scale segmentation algorithm to initially segment the selected data source image, and combining spectral features and texture features to generate the first segmentation result includes:

[0027] Acquire raw remote sensing image data, perform radiometric and geometric corrections on the image quality, and obtain standardized image data;

[0028] Based on the spatial resolution characteristics of the standardized image data, a multi-scale segmentation algorithm is used to set the combination of segmentation parameters and determine the segmentation scale range at different levels.

[0029] Calculate the mean, standard deviation, and spectral index of each band to obtain pixel-level spectral feature vectors;

[0030] The gray-level co-occurrence matrix method is used to perform texture analysis on the image region corresponding to the spectral feature vector to obtain a set of texture feature parameters.

[0031] Based on the spectral feature vector and texture feature parameter set, the K-means clustering algorithm is used to cluster the image pixels to determine the initial segmentation unit;

[0032] The initial segmentation unit is optimized by a boundary detection operator. If the boundary continuity meets a preset threshold, the current segmentation is maintained. If the boundary is discontinuous, the segmentation parameters are adjusted and the segmentation is repeated to obtain an optimized segmentation result.

[0033] The optimized segmentation results are evaluated using a segmentation accuracy evaluation index to obtain the first segmentation result dataset.

[0034] Preferably, the process of optimizing the boundaries of the first segmentation result based on a region growing algorithm, and merging adjacent regions into the same desertification patch if the spectral difference is less than a preset tolerance, includes:

[0035] Obtain the spectral feature vectors of each region in the first segmentation result, and extract the mean spectral feature parameters of each region through spectral reflectance data to obtain the regional spectral feature dataset;

[0036] Based on the regional spectral feature dataset, a neighboring region relationship matrix is ​​constructed. The eight-neighbor connectivity analysis method is used to identify the location information of neighboring regions for each region and to determine the neighboring region pairing relationship table.

[0037] The spectral feature difference values ​​between each region in the adjacent region pairing table are calculated using the Euclidean distance algorithm to obtain the spectral difference value matrix.

[0038] If the difference value of a pair of adjacent regions in the spectral difference matrix is ​​less than a preset tolerance threshold, the region pair is marked as to be merged, and a list of regions to be merged is obtained.

[0039] According to the list of regions to be merged, a region merging operation is performed. A region growing algorithm is used to merge the pixels of the marked regions to the same label value to obtain the merged region segmentation result.

[0040] The boundary contours of the merged region segmentation results are extracted by morphological boundary extraction operators to obtain the boundary coordinate data of desertification patches.

[0041] Based on the boundary coordinate data of the desertification patches, the patch geometry descriptor is reconstructed, the area-to-perimeter ratio and compactness index of each patch are calculated, and the final desertification patch segmentation result is determined.

[0042] Preferably, the process of extracting indicators of construction land expansion and agricultural activity intensity from time-series data and establishing a quantitative relationship model between anthropogenic factors and desertification patch changes includes:

[0043] The expansion of built-up land was obtained using multi-temporal remote sensing imagery, and the intensity value of agricultural areas was calculated using the NDVI interpolation method.

[0044] A feature matrix is ​​constructed based on the number of desertification patches and the intensity of their expansion. The Pearson coefficient is used to screen for significant anthropogenic driving terms. If the correlation is higher than a preset threshold, a multiple linear regression model of desertification dynamics and anthropogenic driving terms is established.

[0045] The significance of the regression terms was tested using residual analysis, and driving variables that failed the t-test were removed.

[0046] The sliding window method was used to segment the time series data, verify the model stability at different time periods, and output the standardized regression coefficients of agricultural intensity value of construction land expansion and desertification patch number.

[0047] Preferably, for data in the missing time period, the process of generating a continuous surface using the Kriging spatiotemporal interpolation algorithm, combined with the spatial distribution of existing data points and the desertification evolution trend, includes:

[0048] Historical time series data of all observation points in the desertification monitoring area are obtained. The distribution of missing data at each observation point in different time periods is identified by a data integrity detection algorithm, and a data missing location index table is obtained.

[0049] Based on the spatial coordinate information in the missing data location index table, the spatial distance matrix between each observation point is calculated. The semi-variogram analysis method is used to quantify the spatial correlation strength of desertification at different distances and determine the optimal spatial correlation parameters.

[0050] The covariance matrix of the Kriging interpolation model is constructed based on the spatial correlation parameters. If the spatial correlation coefficient of the observation point is greater than the preset threshold, the corresponding point is included in the neighborhood range of the interpolation calculation to obtain the effective neighborhood observation point set for each missing data point.

[0051] Based on the desertification observation values ​​in the effective neighborhood observation point set, the desertification state estimate for each missing time period is calculated using the Kriging interpolation model to obtain the completed spatiotemporal data matrix.

[0052] Based on the changes in desertification degree between adjacent time points in the completed spatiotemporal data matrix, the spatiotemporal evolution parameters of each observation point are calculated. If the rate of change between adjacent time points exceeds the preset change threshold, the interpolation results are corrected by time constraints to obtain the final spatiotemporal distribution data of desertification degree.

[0053] Using the final spatiotemporal distribution data of desertification, the bilinear interpolation method is used to generate desertification values ​​for regular grid points within the study area, and a continuous surface model covering the entire monitoring area is constructed.

[0054] Based on the desertification degree value and spatial coordinates of each grid point in the continuous surface model, the boundary lines of different desertification degree levels are extracted by the contour line tracing algorithm to obtain the complete spatial distribution pattern of desertification.

[0055] Preferably, the process of integrating multi-period segmentation results and anthropogenic factor quantification data to output a dynamic change map of desertification boundaries and a weighted distribution map of anthropogenic impacts includes:

[0056] Multi-temporal desertification remote sensing image segmentation results data are obtained, and standardized temporal data sets are obtained through time series alignment processing;

[0057] Based on the segmentation results of each period in the standardized temporal data set, a pixel-level change detection method is used to determine the change status of the desertification boundary and obtain boundary change vector data.

[0058] The boundary change vector data is analyzed. If a pixel undergoes a category change in a continuous time phase, the change intensity value is calculated to obtain the characteristic parameters of the desertification evolution trajectory.

[0059] Based on the characteristic parameters of the desertification evolution trajectory, combined with the preset change threshold conditions, the trend of desertification expansion or reversal is determined, and the dynamic change classification results are obtained.

[0060] Based on the dynamic change classification results, quantitative data of anthropogenic factors at the corresponding spatial locations are obtained, and the correlation strength between anthropogenic factors and desertification change is calculated using correlation analysis methods.

[0061] If the correlation strength exceeds a preset threshold, the corresponding factor is determined to be the dominant influencing factor, and the distribution value of the human influence weight is obtained.

[0062] Based on the weight distribution values ​​and spatial coordinate information, a dynamic change map of desertification boundaries and a spatial distribution map of human influence weights are generated.

[0063] This invention also provides a desertification monitoring system based on multi-source remote sensing data fusion, comprising:

[0064] The data quality assessment module is used to acquire multi-source satellite remote sensing data and use a preset data quality assessment algorithm to perform a weighted score on the spectral resolution, cloud coverage, and temporal resolution of the multi-source satellite remote sensing data to obtain a score result.

[0065] The dynamic allocation module is used to establish a dynamic allocation mechanism based on the scoring results. If the current data score is lower than a preset threshold, it will automatically switch to a backup data source with higher priority.

[0066] The multi-scale segmentation module is used to perform initial segmentation on selected data source images using a multi-scale segmentation algorithm, and generate the first segmentation result by combining spectral features and texture features.

[0067] The boundary optimization module is used to optimize the boundary of the first segmentation result based on the region growing algorithm. If the spectral difference between adjacent regions is less than the preset tolerance, they are merged into the same desertification patch.

[0068] The anthropogenic factor extraction module is used to extract indicators of construction land expansion and agricultural activity intensity from time series data, and to establish a quantitative relationship model between anthropogenic factors and desertification patch changes.

[0069] The spatiotemporal interpolation module is used to generate continuous surfaces for missing time period data by employing the Kriging spatiotemporal interpolation algorithm and combining the spatial distribution of existing data points with the desertification evolution trend.

[0070] The output module is used to integrate multi-period segmentation results and human factor quantification data to output a dynamic change map of desertification boundaries and a weight distribution map of human impact.

[0071] Compared with the prior art, the present invention has the following advantages and technical effects:

[0072] This invention optimizes data source selection through data quality assessment and dynamic allocation. Multi-scale segmentation and region growing algorithms are used to segment images, and desertification patches are extracted by combining spectral and texture features. Anthropogenic factor indicators are extracted from time-series data to establish a quantitative relationship model with desertification change. To address data gaps, a Kriging spatiotemporal interpolation algorithm is used to generate continuous surfaces. Finally, by fusing multi-period segmentation results and anthropogenic factor data, a dynamic change map of desertification boundaries and a weighted distribution map of anthropogenic impacts are output. This invention achieves precise monitoring of desertification processes and quantitative assessment of anthropogenic impacts, providing a scientific basis for desertification control. Attached Figure Description

[0073] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings:

[0074] Figure 1 This is a schematic diagram of the method flow according to an embodiment of the present invention;

[0075] Figure 2 This is a schematic diagram of the system structure according to an embodiment of the present invention. Detailed Implementation

[0076] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0077] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0078] Example 1

[0079] like Figure 1 As shown, this embodiment provides a desertification monitoring method based on multi-source remote sensing data fusion, including:

[0080] Acquire multi-source satellite remote sensing data, and use a preset data quality assessment algorithm to weight and score the spectral resolution, cloud coverage, and temporal resolution of the multi-source satellite remote sensing data to obtain the scoring results;

[0081] A dynamic allocation mechanism is established based on the scoring results. If the current data score is lower than the preset threshold, it will automatically switch to a backup data source with higher priority.

[0082] A multi-scale segmentation algorithm is used to perform initial segmentation on images from selected data sources, and the first segmentation result is generated by combining spectral features and texture features.

[0083] The boundary of the first segmentation result is optimized based on the region growing algorithm. If the spectral difference between adjacent regions is less than the preset tolerance, they are merged into the same desertification patch.

[0084] We extract indicators of construction land expansion and agricultural activity intensity from time-series data to establish a quantitative relationship model between human factors and desertification patch changes;

[0085] For data in the missing time period, the Kriging spatiotemporal interpolation algorithm is used to generate a continuous surface by combining the spatial distribution of existing data points and the desertification evolution trend.

[0086] By integrating multi-period segmentation results and anthropogenic factor quantification data, a dynamic change map of desertification boundaries and a weight distribution map of anthropogenic influences are output.

[0087] Furthermore, the process of acquiring multi-source satellite remote sensing data and using a pre-defined data quality assessment algorithm to weight and score the multi-source satellite remote sensing data to obtain the scoring results includes:

[0088] Acquire spectral, cloud cover, and temporal values ​​from multi-source satellite data, and calculate spectral resolution, cloud coverage, and temporal resolution using a pre-defined evaluation method;

[0089] The spectral resolution, cloud cover, and temporal resolution are normalized according to the weight values. If the cloud cover value exceeds the preset value, the corresponding data source is removed.

[0090] The normalized indicators are combined using a weighted method to obtain a score, and the score and the corresponding multi-source data from the data source are stored.

[0091] In one possible implementation, data from the Landsat-8 satellite's OLI sensor is used as an example when acquiring spectral values ​​from multi-source satellite data. This sensor provides spectral information across nine bands, from visible light to shortwave infrared, with bands 2 (blue light) and 4 (red light) commonly used for vegetation index calculations. The spectral resolution is calculated using a pre-defined evaluation method, comparing the bandwidth differences between bands. For example, band 1 has a bandwidth of 0.43-0.45 μm, and its narrowband characteristics give it higher weight in desertification monitoring.

[0092] Specifically, cloud cover processing uses Sentinel-2 data as an example, whose metadata includes the percentage of cloud coverage for each image. When the preset cloud cover threshold is 20%, if the cloud cover in a particular image reaches 35%, a removal mechanism is triggered. This design effectively avoids interference from clouds in the calculation of surface reflectance, improving the reliability of subsequent data fusion. Temporal resolution evaluation needs to be combined with the satellite revisit cycle. For example, MODIS data covers the same area daily, while Gaofen-7 requires 3 days; in dynamic monitoring scenarios, the former has a higher weight.

[0093] Preferably, the normalization process employs a linear transformation method. For example, mapping the spectral resolution from 0-100nm to the 0-1 interval, a satellite's 30nm bandwidth, after conversion, yields 0.7 points. When cloud cover is reverse-normalized, 10% cloud cover can be converted to 0.9 points. Temporal resolution is measured in days, with a 1-day revisit cycle yielding 1 point and a 3-day cycle yielding 0.33 points. After weighting the three indicators with weights of 0.4, 0.3, and 0.3 respectively, the final score is 0.7×0.4 + 0.9×0.3 + 1×0.3 = 0.85 points.

[0094] Understandably, this scheme filters low-quality data through dynamic thresholds and combines multi-dimensional quantitative evaluation to ensure that the fusion results favor data sources with rich spectral detail, minimal cloud interference, and high timeliness. The storage stage employs a hierarchical structure, associating the scoring values ​​with the corresponding data source's metadata and original image paths for easy subsequent traceability and reuse.

[0095] Furthermore, a dynamic allocation mechanism is established based on the scoring results. If the current data score is lower than a preset threshold, the process of automatically switching to a higher-priority backup data source includes:

[0096] Obtain real-time performance metrics data from the data source; collect key parameters such as data transmission latency, connection stability, and response time through the monitoring system to obtain comprehensive performance status information of the current data source.

[0097] Based on the comprehensive performance status information, a weighted average algorithm is used to calculate the real-time score of the data source, thereby obtaining a standardized data source quality score.

[0098] Determine if the real-time score is below a preset threshold. If so, automatically trigger the backup resource activation mechanism to obtain available resource information from a pre-configured backup data source list and determine the current status of the backup resources.

[0099] Available backup resources are sorted using a priority ranking algorithm. Priority scores are calculated based on resource capacity, historical stability, and geographical location to obtain a priority sequence of backup resources.

[0100] Select the highest-ranked backup data source as the switching target based on the priority order, establish a connection channel with the target data source, and obtain connection status confirmation information;

[0101] If the connection status confirmation information shows that the connection is successful, a data stream switching operation is performed to redirect the business request from the original data source to the target backup data source, thus completing the dynamic switching process of the data source.

[0102] A continuous monitoring mechanism is used to track the performance of the switched data source, record the execution time of the switching operation, data integrity and business continuity, determine the success status of the switching operation and update the system operation log.

[0103] For example, the system first establishes a multi-dimensional scoring model and uses a weighted average algorithm to calculate the comprehensive score of the data source, where the data accuracy weight is 0.4, the response time weight is 0.3, and the availability weight is 0.3. When the main data source A has an accuracy score of 85, a response time score of 70, and an availability score of 60 at a certain moment, the comprehensive score is 85×0.4+70×0.3+60×0.3=73.

[0104] The system sets a preset threshold of 75 points. If the current score of 73 points is lower than the threshold, an automatic switchover mechanism is triggered. Backup data sources are prioritized, with backup source B having priority 1 and backup source C having priority 2. The system uses a sliding window algorithm to perform real-time health checks on backup source B, detecting an average response time of 120 milliseconds and a success rate of 98.5% over the past 5 minutes. Once the switchover conditions are met, a seamless switchover is performed. During the switchover process, a dual-write strategy is used to ensure data consistency, and a circuit breaker mechanism is activated with a 60-second cooldown to prevent frequent switchovers. The system continuously monitors the recovery status of the primary data source A. When its score exceeds 80 points for three consecutive checks and the duration reaches 300 seconds, an automatic switchback operation is performed. The entire process achieves millisecond-level response through an event-driven architecture, ensuring business continuity and data service quality.

[0105] Furthermore, the process of using a multi-scale segmentation algorithm to initially segment the selected data source image, and combining spectral and texture features to generate the first segmentation result includes:

[0106] Acquire raw remote sensing image data, perform radiometric and geometric corrections on the image quality, and obtain standardized image data;

[0107] Based on the spatial resolution characteristics of standardized image data, a multi-scale segmentation algorithm is used to set the combination of segmentation parameters and determine the segmentation scale range at different levels.

[0108] Calculate the mean, standard deviation, and spectral index of each band to obtain pixel-level spectral feature vectors;

[0109] The gray-level co-occurrence matrix method is used to perform texture analysis on the image region corresponding to the spectral feature vector to obtain a set of texture feature parameters.

[0110] Based on the spectral feature vector and texture feature parameter set, the K-means clustering algorithm is used to cluster the image pixels to determine the initial segmentation unit;

[0111] The initial segmentation unit is optimized by a boundary detection operator. If the boundary continuity meets the preset threshold, the current segmentation is maintained. If the boundary is discontinuous, the segmentation parameters are adjusted and the segmentation is repeated to obtain the optimized segmentation result.

[0112] The quality of the optimized segmentation results is evaluated using a segmentation accuracy evaluation index, resulting in the first segmentation result dataset.

[0113] For example, the high-resolution remote sensing image is first preprocessed, using a Gaussian filter for noise removal, with a filter window size of 3×3 and a standard deviation of 0.8. Then, a multi-scale segmentation algorithm is implemented, selecting three segmentation scales of 50, 100, and 150, with a shape factor of 0.3 and a compactness factor of 0.5. Similar regions are merged pixel-by-pixel using a region growing algorithm. In the spectral feature extraction stage, the mean and standard deviation of the red, green, blue, and near-infrared bands for each segmented object are calculated, along with the Normalized Difference Vegetation Index (NDVI). The NDVI value for vegetation areas is typically greater than 0.4, while that for water bodies is less than 0.1. Texture feature extraction uses the gray-level co-occurrence matrix method, with a window size of 7×7. Contrast, correlation, energy, and entropy values ​​are calculated at four angles: 0°, 45°, 90°, and 135°. The contrast value for building areas is typically between 15 and 25, while the entropy value for farmland areas is generally between 8 and 12. The weight of spectral features was set to 0.6 and the weight of texture features was set to 0.4. Weighted Euclidean distance was used to calculate the similarity between objects. When the similarity threshold was less than 2.5, objects were merged. Finally, an initial segmentation result containing attribute information such as land cover category, boundary coordinates, and area size was generated, providing basic data support for subsequent accurate classification.

[0114] Furthermore, based on the region growing algorithm, the boundary of the first segmentation result is optimized. If the spectral difference between adjacent regions is less than the preset tolerance, the process of merging them into the same desertification patch includes:

[0115] Obtain the spectral feature vectors of each region in the first segmentation result, and extract the mean spectral feature parameters of each region through spectral reflectance data to obtain the regional spectral feature dataset;

[0116] Based on the regional spectral feature dataset, a neighboring region relationship matrix is ​​constructed. The eight-neighbor connectivity analysis method is used to identify the location information of neighboring regions for each region and determine the neighboring region pairing relationship table.

[0117] The spectral feature difference values ​​between each region in the adjacent region pairing table are calculated using the Euclidean distance algorithm to obtain the spectral difference value matrix.

[0118] If the difference value of a pair of adjacent regions in the spectral difference matrix is ​​less than the preset tolerance threshold, the region pair is marked as to be merged, and a list of regions to be merged is obtained.

[0119] Based on the list of regions to be merged, a region merging operation is performed. A region growing algorithm is used to merge the pixels of the marked regions to the same label value, resulting in the merged region segmentation result.

[0120] The boundary contours of the merged region segmentation results are extracted by morphological boundary extraction operators to obtain the boundary coordinate data of desertification patches.

[0121] Based on the boundary coordinate data of desertification patches, the geometric descriptors of the patches are reconstructed, the area-to-perimeter ratio and compactness index of each patch are calculated, and the final desertification patch segmentation results are determined.

[0122] For example, boundary optimization based on region growing algorithms first requires establishing a seed point selection mechanism. By calculating the spectral gradient value of each pixel, pixels with gradient values ​​less than 0.05 are selected as initial seed points. These points are typically located in the center of regions with relatively uniform spectra. Next, the region growing process is executed, expanding outwards from the seed points in an eight-neighbor direction. The Euclidean distance between adjacent pixels and the average spectral value of the current region is calculated. When the distance is less than a preset threshold of 0.08, the pixel is included in the current region. During the growing process, the average spectral feature value of the region is updated in real time to ensure spectral consistency within the region. After all pixels have been assigned to regions, an adjacent region merging algorithm is initiated. The spectral angular distance and spectral information divergence between adjacent regions are calculated. If the spectral angle is less than 15 degrees and the information divergence is less than 0.12, they are determined to be homogeneous regions and a merging operation is performed. The merging process uses a weighted average method to recalculate the spectral features of the new region, with the weights determined based on the number of pixels in each sub-region. Finally, morphological opening and closing operations were applied to eliminate small noise patches. The minimum patch area threshold was set to 9 pixels. Patches smaller than this threshold were merged into the adjacent most similar regions. Through this series of optimization processes, the accuracy of desertification patch boundaries was improved by about 18%, effectively reducing oversegmentation.

[0123] Furthermore, the process of extracting indicators of construction land expansion and agricultural activity intensity from time-series data to establish a quantitative model of the relationship between anthropogenic factors and desertification patch changes includes:

[0124] The expansion of built-up land was obtained using multi-temporal remote sensing imagery, and the intensity value of agricultural areas was calculated using the NDVI interpolation method.

[0125] A feature matrix is ​​constructed based on the number of desertification patches and the intensity of their expansion. The Pearson coefficient is used to screen for significant anthropogenic driving terms. If the correlation is higher than a preset threshold, a multiple linear regression model of desertification dynamics and anthropogenic driving terms is established.

[0126] The significance of the regression terms was tested using residual analysis, and driving variables that failed the t-test were removed.

[0127] The sliding window method was used to segment the time series data, verify the model stability at different time periods, and output the standardized regression coefficients of agricultural intensity value of construction land expansion and desertification patch number.

[0128] For example, in desertification monitoring, multi-temporal remote sensing images can be used to extract the amount of construction land expansion.

[0129] For example, Landsat images from 2015 and 2020 are selected, and the construction land area of ​​the two periods is extracted through supervised classification. After overlay analysis, the expansion area is calculated. If the construction land in a certain area increases from 50 hectares to 80 hectares, the expansion amount is 30 hectares. The expansion intensity can be quantified by the rate of change of area per unit time.

[0130] In one possible implementation, the NDVI interpolation method is used to assess the intensity of agricultural zones.

[0131] For example, by selecting growing season images and calculating the average NDVI over two periods, if the NDVI of a certain farmland area decreases from 0.6 to 0.4, the difference of 0.2 reflects the change in the intensity of agricultural activities. Combined with the number of desertification patches (e.g., increasing from 15 to 25), a feature matrix containing the expansion amount, NDVI difference, and patch number can be constructed.

[0132] Specifically, when using the Pearson coefficient to screen for significant human-driven factors, a preset threshold of 0.7 is set. If the correlation coefficient between the amount of construction land expansion and the number of patches is 0.8, which is higher than the threshold, it is considered a strongly correlated driving factor. Conversely, if the correlation coefficient between agricultural intensity and the number of patches is 0.5, it is excluded.

[0133] In one embodiment, after the multiple linear regression model is established, residual analysis is used to test the significance of the driving terms.

[0134] For example, the regression coefficient for the expansion of construction land is 0.3 and passes the t-test, indicating that for every 1 unit increase in construction land, the number of desertification patches is expected to increase by 0.3. However, the agricultural intensity coefficient fails the test and is therefore removed from the model.

[0135] Understandably, the sliding window method can be used to verify the temporal stability of the model.

[0136] For example, if the data from 2010 to 2020 is divided into 5-year windows, and the fluctuation of the construction land expansion coefficient in each period is less than 10%, it indicates that the model is robust. The final output standardized regression coefficients can directly compare the contribution of different driving factors.

[0137] Furthermore, for data in the missing time periods, the process of generating a continuous surface using the Kriging spatiotemporal interpolation algorithm, combined with the spatial distribution of existing data points and the desertification evolution trend, includes:

[0138] Historical time series data of all observation points in the desertification monitoring area are obtained. The distribution of missing data at each observation point in different time periods is identified by a data integrity detection algorithm, and a data missing location index table is obtained.

[0139] Based on the spatial coordinate information in the missing data location index table, the spatial distance matrix between each observation point is calculated. The semi-variogram analysis method is used to quantify the spatial correlation strength of desertification at different distances and determine the optimal spatial correlation parameters.

[0140] The covariance matrix of the Kriging interpolation model is constructed based on the spatial correlation parameters. If the spatial correlation coefficient of the observation point is greater than the preset threshold, the corresponding point is included in the neighborhood range of the interpolation calculation to obtain the effective neighborhood observation point set for each missing data point.

[0141] Based on the desertification observations in the effective neighborhood observation point set, the desertification state estimates for each missing time period are calculated using the Kriging interpolation model, resulting in a completed spatiotemporal data matrix.

[0142] Based on the changes in desertification degree between adjacent time points in the completed spatiotemporal data matrix, the spatiotemporal evolution parameters of each observation point are calculated. If the rate of change between adjacent time points exceeds the preset change threshold, the interpolation results are corrected by time constraints to obtain the final spatiotemporal distribution data of desertification degree.

[0143] Using the final spatiotemporal distribution data of desertification, the bilinear interpolation method is used to generate desertification values ​​for regular grid points within the study area, and a continuous surface model covering the entire monitoring area is constructed.

[0144] Based on the desertification degree value and spatial coordinates of each grid point in the continuous surface model, the boundary lines of different desertification degree levels are extracted by the contour line tracing algorithm to obtain the complete spatial distribution pattern of desertification.

[0145] For example, in a data integrity check for desertification monitoring, 100 observation points were set up in a study area. A historical data scan revealed missing data at 15 points between 2018 and 2022. Specifically, observation point A lacked data for three months from March to May 2019, and observation point B lacked data from July to September 2020. A data missing location index table recorded the latitude and longitude coordinates, missing time period, and missing data type for each missing point, providing a precise spatial positioning basis for subsequent interpolation.

[0146] In one possible implementation, the spatial distance matrix is ​​calculated using the Euclidean distance method. Taking observation point A as an example, its distances to its eight neighboring points vary from 500 meters, 800 meters, 1200 meters to 3000 meters. Semivariogram analysis shows that when the distance is less than 1500 meters, the spatial correlation coefficient of desertification remains above 0.75, but after exceeding 1500 meters, the correlation rapidly decays to below 0.4. Therefore, 1500 meters is determined to be the optimal spatial correlation parameter, which directly affects the accuracy and reliability of Kriging interpolation.

[0147] Specifically, in the construction of the covariance matrix of the Kriging interpolation model, a spatial correlation threshold of 0.6 was preset. For observation point A with missing data, six valid observation points were identified within a 1500-meter neighborhood, with correlation coefficients of 0.82, 0.76, 0.71, 0.68, 0.65, and 0.63, all exceeding the threshold requirement. The desertification levels of these six neighborhood points were 0.3 for mild desertification, 0.6 for moderate desertification, 0.4 for mild desertification, 0.5 for moderate desertification, 0.35 for mild desertification, and 0.55 for moderate desertification. Through weight allocation calculation, observation points that are closer to the data received higher weights, and the final interpolation yielded an estimated desertification level of 0.48 for observation point A during the missing period.

[0148] In one embodiment, the calculation of spatiotemporal evolution parameters is based on the analysis of the rate of change between adjacent time points. At a certain observation point, the desertification level was 0.4 in April 2019 and 0.7 in May, with a rate of change reaching 0.3, exceeding the preset change threshold of 0.25. The system automatically triggers a time constraint correction mechanism, combining the historical change trend of this location with the concurrent changes of surrounding locations, adjusting the interpolation result for May to 0.55 to ensure the continuity and rationality of the time series.

[0149] For example, the bilinear interpolation method generates a 50m × 50m regular grid within the study area, totaling 40,000 grid points. The desertification level value for each grid point is obtained by weighted averaging of the four nearest surrounding observation points. The continuous surface model shows a clear spatial gradient distribution of desertification level across the entire monitoring area, with less desertification in the southeast and relatively more severe desertification in the northwest.

[0150] Understandably, the contour tracking algorithm sets three contour thresholds of 0.3, 0.5, and 0.7 based on the desertification severity classification standard, corresponding to the boundaries of light, moderate, and severe desertification, respectively. The algorithm automatically identifies and extracts these contour lines, forming a clear spatial distribution pattern map of desertification. Lightly desertified areas account for 45% of the total area, moderately desertified areas account for 35%, and severely desertified areas account for 20%, providing a precise spatial location and area statistics basis for desertification control.

[0151] Furthermore, the process of integrating multi-period segmentation results and anthropogenic factor quantification data to output a dynamic change map of desertification boundaries and a weighted distribution map of anthropogenic impacts includes:

[0152] Multi-temporal desertification remote sensing image segmentation results data are obtained, and standardized temporal data sets are obtained through time series alignment processing;

[0153] Based on the segmentation results of each period in the standardized temporal data set, a pixel-level change detection method is used to determine the change status of the desertification boundary and obtain boundary change vector data.

[0154] By analyzing the boundary change vector data, if a pixel undergoes a category change in a continuous time phase, the change intensity value is calculated to obtain the characteristic parameters of the desertification evolution trajectory.

[0155] Based on the characteristic parameters of desertification evolution trajectory and combined with the preset change threshold conditions, the trend of desertification expansion or reversal is determined, and the dynamic change classification results are obtained.

[0156] Based on the dynamic change classification results, quantitative data of anthropogenic factors at the corresponding spatial locations are obtained, and the correlation strength between anthropogenic factors and desertification change is calculated through correlation analysis.

[0157] If the correlation strength exceeds the preset threshold, the corresponding factor is determined to be the dominant influencing factor, and the distribution value of the human influence weight is obtained.

[0158] Based on the weight distribution values ​​and spatial coordinate information, a dynamic change map of desertification boundaries and a spatial distribution map of human influence weights are generated.

[0159] For example, firstly, the desertification segmentation results for 2010, 2015, and 2020 are processed using time series analysis algorithms. A change detection matrix is ​​used to calculate the migration probability of each pixel category, with a threshold of 0.3 for the probability of grassland migrating to desertification. When the migration probability exceeds this threshold for two consecutive periods, the area is marked as a desertification expansion zone. Simultaneously, a quantitative model of anthropogenic factors is established, standardizing factors such as population density, road density, and agricultural development intensity. The weighting coefficients are determined using the analytic hierarchy process (AHP) to be 0.4, 0.3, and 0.3, respectively, and a weighted summation is used to obtain the comprehensive anthropogenic impact index. Next, spatial overlay analysis is used to fuse desertification boundary change data with anthropogenic factor data. A buffer zone analysis method is used to establish a 500-meter buffer zone on both sides of the desertification boundary, and the spatial distribution characteristics of the anthropogenic impact index within the buffer zone are statistically analyzed. The spatial clustering degree of anthropogenic impact is calculated using kernel density estimation, with a search radius of 1000 meters. A kernel density value greater than 0.8 is considered to indicate significant anthropogenic impact. Finally, a continuous surface of the dynamic changes in desertification boundary is generated using the spatial interpolation algorithm of the geographic information system. The inverse distance weighted interpolation method is adopted, with the power parameter set to 2 and the search neighborhood set to 12 nearest points. At the same time, the natural discontinuity classification method is used to divide the human influence weight into 5 levels, generating a weight distribution map to realize the visualization of the spatial and temporal evolution process of desertification and human driving factors.

[0160] Example 2

[0161] Based on the same inventive concept, this embodiment also provides a desertification monitoring system based on multi-source remote sensing data fusion, including:

[0162] The data quality assessment module is used to acquire multi-source satellite remote sensing data. It uses a preset data quality assessment algorithm to weight and score the spectral resolution, cloud coverage, and temporal resolution of the multi-source satellite remote sensing data to obtain the score results.

[0163] The dynamic allocation module is used to establish a dynamic allocation mechanism based on the scoring results. If the current data score is lower than the preset threshold, it will automatically switch to a backup data source with higher priority.

[0164] The multi-scale segmentation module is used to perform initial segmentation on selected data source images using a multi-scale segmentation algorithm, and generate the first segmentation result by combining spectral features and texture features.

[0165] The boundary optimization module is used to optimize the boundary of the first segmentation result based on the region growing algorithm. If the spectral difference between adjacent regions is less than the preset tolerance, they are merged into the same desertification patch.

[0166] The anthropogenic factor extraction module is used to extract indicators of construction land expansion and agricultural activity intensity from time series data, and to establish a quantitative relationship model between anthropogenic factors and desertification patch changes.

[0167] The spatiotemporal interpolation module is used to generate continuous surfaces for missing time period data by employing the Kriging spatiotemporal interpolation algorithm and combining the spatial distribution of existing data points with the desertification evolution trend.

[0168] The output module is used to integrate multi-period segmentation results and human factor quantification data to output a dynamic change map of desertification boundaries and a weight distribution map of human impact.

[0169] The desertification monitoring system based on multi-source remote sensing data fusion provided in this embodiment has all the advantages of the desertification monitoring method based on multi-source remote sensing data fusion provided in Embodiment 1.

[0170] The above are merely preferred embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A desertification monitoring method based on multi-source remote sensing data fusion, characterized in that, include: Acquire multi-source satellite remote sensing data, and use a preset data quality assessment algorithm to weight and score the spectral resolution, cloud coverage, and temporal resolution of the multi-source satellite remote sensing data to obtain the scoring results; A dynamic allocation mechanism is established based on the scoring results. If the current data score is lower than a preset threshold, it will automatically switch to a backup data source with higher priority. A multi-scale segmentation algorithm is used to perform initial segmentation on images from selected data sources, and the first segmentation result is generated by combining spectral features and texture features. The first segmentation result is optimized based on the region growing algorithm. If the spectral difference between adjacent regions is less than the preset tolerance, they are merged into the same desertification patch. We extract indicators of construction land expansion and agricultural activity intensity from time-series data to establish a quantitative relationship model between human factors and desertification patch changes; For data in the missing time period, the Kriging spatiotemporal interpolation algorithm is used to generate a continuous surface by combining the spatial distribution of existing data points and the desertification evolution trend. By integrating multi-period segmentation results and anthropogenic factor quantification data, a dynamic change map of desertification boundaries and a weight distribution map of anthropogenic influences are output.

2. The method according to claim 1, characterized in that, The process of acquiring multi-source satellite remote sensing data, weighting and scoring the multi-source satellite remote sensing data using a preset data quality assessment algorithm, and obtaining the scoring results includes: Acquire spectral, cloud cover, and temporal values ​​from multi-source satellite data, and calculate spectral resolution, cloud coverage, and temporal resolution using a pre-defined evaluation method; The spectral resolution, cloud coverage, and temporal resolution are normalized according to the weight values. If the cloud cover value exceeds the preset value, the corresponding data source is removed. The normalized indicators are combined using a weighted method to obtain a score value, and the score value and the corresponding multi-source data from the data source are stored.

3. The method according to claim 1, characterized in that, Based on the scoring results, a dynamic allocation mechanism is established. If the current data score is lower than a preset threshold, the process of automatically switching to a higher-priority backup data source includes: Obtain real-time performance metrics data from the data source; collect key parameters such as data transmission latency, connection stability, and response time through the monitoring system to obtain comprehensive performance status information of the current data source. Based on the comprehensive performance status information, a weighted average algorithm is used to calculate the real-time score of the data source, thereby obtaining a standardized data source quality score. Determine if the real-time score is below a preset threshold. If so, automatically trigger the backup resource activation mechanism to obtain available resource information from a pre-configured backup data source list and determine the current status of the backup resources. Available backup resources are sorted using a priority ranking algorithm. Priority scores are calculated based on resource capacity, historical stability, and geographical location to obtain a priority sequence of backup resources. Select the highest-ranked backup data source as the switching target according to the priority sequence, establish a connection channel with the target data source, and obtain connection status confirmation information. If the connection status confirmation information shows that the connection is successful, a data stream switching operation is performed to redirect the business request from the original data source to the target backup data source, thus completing the dynamic switching process of the data source. A continuous monitoring mechanism is used to track the performance of the switched data source, record the execution time of the switching operation, data integrity and business continuity, determine the success status of the switching operation and update the system operation log.

4. The method according to claim 1, characterized in that, The process of using a multi-scale segmentation algorithm to initially segment selected data source images and then combining spectral and texture features to generate the first segmentation result includes: Acquire raw remote sensing image data, perform radiometric and geometric corrections on the image quality, and obtain standardized image data; Based on the spatial resolution characteristics of the standardized image data, a multi-scale segmentation algorithm is used to set the combination of segmentation parameters and determine the segmentation scale range at different levels. Calculate the mean, standard deviation, and spectral index of each band to obtain pixel-level spectral feature vectors; The gray-level co-occurrence matrix method is used to perform texture analysis on the image region corresponding to the spectral feature vector to obtain a set of texture feature parameters. Based on the spectral feature vector and texture feature parameter set, the K-means clustering algorithm is used to cluster the image pixels to determine the initial segmentation unit; The initial segmentation unit is optimized by a boundary detection operator. If the boundary continuity meets a preset threshold, the current segmentation is maintained. If the boundary is discontinuous, the segmentation parameters are adjusted and the segmentation is repeated to obtain an optimized segmentation result. The optimized segmentation results are evaluated using a segmentation accuracy evaluation index to obtain the first segmentation result dataset.

5. The method according to claim 1, characterized in that, The process of merging adjacent regions into the same desertification patch based on the boundary optimization of the first segmentation result using a region growing algorithm includes: Obtain the spectral feature vectors of each region in the first segmentation result, and extract the mean spectral feature parameters of each region through spectral reflectance data to obtain the regional spectral feature dataset; Based on the regional spectral feature dataset, a neighboring region relationship matrix is ​​constructed. The eight-neighbor connectivity analysis method is used to identify the location information of neighboring regions for each region and to determine the neighboring region pairing relationship table. The spectral feature difference values ​​between each region in the adjacent region pairing table are calculated using the Euclidean distance algorithm to obtain the spectral difference value matrix. If the difference value of a pair of adjacent regions in the spectral difference matrix is ​​less than a preset tolerance threshold, the region pair is marked as to be merged, and a list of regions to be merged is obtained. According to the list of regions to be merged, a region merging operation is performed. A region growing algorithm is used to merge the pixels of the marked regions to the same label value to obtain the merged region segmentation result. The boundary contours of the merged region segmentation results are extracted by morphological boundary extraction operators to obtain the boundary coordinate data of desertification patches. Based on the boundary coordinate data of the desertification patches, the patch geometry descriptor is reconstructed, the area-to-perimeter ratio and compactness index of each patch are calculated, and the final desertification patch segmentation result is determined.

6. The method according to claim 1, characterized in that, The process of extracting indicators of construction land expansion and agricultural activity intensity from time-series data to establish a quantitative model of the relationship between anthropogenic factors and desertification patch changes includes: The expansion of built-up land was obtained using multi-temporal remote sensing imagery, and the intensity value of agricultural areas was calculated using the NDVI interpolation method. A feature matrix is ​​constructed based on the number of desertification patches and the intensity of their expansion. The Pearson coefficient is used to screen for significant anthropogenic driving terms. If the correlation is higher than a preset threshold, a multiple linear regression model of desertification dynamics and anthropogenic driving terms is established. The significance of the regression terms was tested using residual analysis, and driving variables that failed the t-test were removed. The sliding window method was used to segment the time series data, verify the model stability at different time periods, and output the standardized regression coefficients of agricultural intensity value of construction land expansion and desertification patch number.

7. The method according to claim 1, characterized in that, For data in missing time periods, the process of generating a continuous surface using the Kriging spatiotemporal interpolation algorithm, combined with the spatial distribution of existing data points and the trend of desertification evolution, includes: Historical time series data of all observation points in the desertification monitoring area are obtained. The distribution of missing data at each observation point in different time periods is identified by a data integrity detection algorithm, and a data missing location index table is obtained. Based on the spatial coordinate information in the missing data location index table, the spatial distance matrix between each observation point is calculated. The semi-variogram analysis method is used to quantify the spatial correlation strength of desertification at different distances and determine the optimal spatial correlation parameters. The covariance matrix of the Kriging interpolation model is constructed based on the spatial correlation parameters. If the spatial correlation coefficient of the observation point is greater than the preset threshold, the corresponding point is included in the neighborhood range of the interpolation calculation to obtain the effective neighborhood observation point set for each missing data point. Based on the desertification observation values ​​in the effective neighborhood observation point set, the desertification state estimate for each missing time period is calculated using the Kriging interpolation model to obtain the completed spatiotemporal data matrix. Based on the changes in desertification degree between adjacent time points in the completed spatiotemporal data matrix, the spatiotemporal evolution parameters of each observation point are calculated. If the rate of change between adjacent time points exceeds the preset change threshold, the interpolation results are corrected by time constraints to obtain the final spatiotemporal distribution data of desertification degree. Using the final spatiotemporal distribution data of desertification, the bilinear interpolation method is used to generate desertification values ​​for regular grid points within the study area, and a continuous surface model covering the entire monitoring area is constructed. Based on the desertification degree value and spatial coordinates of each grid point in the continuous surface model, the boundary lines of different desertification degree levels are extracted by the contour line tracing algorithm to obtain the complete spatial distribution pattern of desertification.

8. The method according to claim 1, characterized in that, The process of integrating multi-period segmentation results and anthropogenic factor quantification data to output a dynamic change map of desertification boundaries and a weighted distribution map of anthropogenic impacts includes: Multi-temporal desertification remote sensing image segmentation results data are obtained, and standardized temporal data sets are obtained through time series alignment processing; Based on the segmentation results of each period in the standardized temporal data set, a pixel-level change detection method is used to determine the change status of the desertification boundary and obtain boundary change vector data. The boundary change vector data is analyzed. If a pixel undergoes a category change in a continuous time phase, the change intensity value is calculated to obtain the characteristic parameters of the desertification evolution trajectory. Based on the characteristic parameters of the desertification evolution trajectory, combined with the preset change threshold conditions, the trend of desertification expansion or reversal is determined, and the dynamic change classification results are obtained. Based on the dynamic change classification results, quantitative data of anthropogenic factors at the corresponding spatial locations are obtained, and the correlation strength between anthropogenic factors and desertification change is calculated using correlation analysis methods. If the correlation strength exceeds a preset threshold, the corresponding factor is determined to be the dominant influencing factor, and the distribution value of the human influence weight is obtained. Based on the weight distribution values ​​and spatial coordinate information, a dynamic change map of desertification boundaries and a spatial distribution map of human influence weights are generated.

9. A desertification monitoring system based on multi-source remote sensing data fusion, characterized in that, include: The data quality assessment module is used to acquire multi-source satellite remote sensing data and use a preset data quality assessment algorithm to perform a weighted score on the spectral resolution, cloud coverage, and temporal resolution of the multi-source satellite remote sensing data to obtain a score result. The dynamic allocation module is used to establish a dynamic allocation mechanism based on the scoring results. If the current data score is lower than a preset threshold, it will automatically switch to a backup data source with higher priority. The multi-scale segmentation module is used to perform initial segmentation on selected data source images using a multi-scale segmentation algorithm, and generate the first segmentation result by combining spectral features and texture features. The boundary optimization module is used to optimize the boundary of the first segmentation result based on the region growing algorithm. If the spectral difference between adjacent regions is less than the preset tolerance, they are merged into the same desertification patch. The anthropogenic factor extraction module is used to extract indicators of construction land expansion and agricultural activity intensity from time series data, and to establish a quantitative relationship model between anthropogenic factors and desertification patch changes. The spatiotemporal interpolation module is used to generate continuous surfaces for missing time period data by employing the Kriging spatiotemporal interpolation algorithm and combining the spatial distribution of existing data points with the desertification evolution trend. The output module is used to integrate multi-period segmentation results and human factor quantification data to output a dynamic change map of desertification boundaries and a weight distribution map of human impact.

Images