A grassland degradation monitoring method based on remote sensing monitoring

Abstract

The application discloses a grassland degradation monitoring method based on remote sensing monitoring, comprising: acquiring multi-source remote sensing data; preprocessing the multi-source remote sensing data to obtain preprocessed remote sensing data; extracting multi-dimensional feature parameters based on the preprocessed remote sensing data to obtain a multi-dimensional feature parameter set; performing time sequence analysis on vegetation dynamic characteristics, soil dynamic characteristics and human disturbance characteristics by combining multi-time observation data through the multi-dimensional feature parameter set to construct a dynamic characteristic time sequence data set; performing deep fusion on spatial features and time sequence changes in the time sequence by using a convolutional neural network model according to the dynamic characteristic time sequence data set to generate a spatio-temporal fusion feature set; obtaining a degradation grade division result map according to the spatio-temporal fusion feature set; integrating multi-source remote sensing data based on the degradation grade division result map to obtain a refined grassland degradation evaluation result, and monitoring grassland degradation according to the evaluation result.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of remote sensing monitoring, and particularly relates to a grassland degradation monitoring method based on remote sensing monitoring. BACKGROUND

[0002] The health and sustainable development of grassland ecosystems are of critical importance to maintaining global ecological balance, supporting pastoral economies, and addressing climate change. Grassland degradation not only affects ecological functions, but also threatens regional economies and food security. However, existing methods for monitoring grassland degradation often rely on a single data source or simple indicators, making it difficult to capture the complex dynamic changes of grassland ecosystems, especially in complex terrain, human activity interference, and multi-seasonal scenarios. The monitoring accuracy and spatial and temporal coverage are insufficient. This results in a lack of comprehensiveness and reliability in degradation assessment results, making it difficult to meet the needs of scientific management and precise restoration.

[0003] In grassland degradation monitoring, the core challenge is how to effectively integrate multi-source remote sensing data and extract feature parameters that can reflect the dynamic changes of grassland ecology. Different remote sensing data (such as optical images, radar images, and thermal infrared images) have their own physical characteristics, with significant differences in resolution and observation period. The radiation consistency and terrain adaptability of collaborative processing of these data become the primary problem. Relying solely on a single data source or traditional processing methods cannot accurately capture the comprehensive characteristics of grassland vegetation, soil, and human disturbance. The spatio-temporal heterogeneity of these characteristics further exacerbates the problem, and static or single feature extraction cannot reflect the dynamic evolution of continuous growing seasons, resulting in a lack of refinement and dynamization in degradation grade classification.

[0004] Therefore, how to integrate multi-source and multi-temporal remote sensing data, extract dynamic features reflecting grassland vegetation, soil, and human disturbance, and achieve the fusion of spatio-temporal features and the accurate classification of degradation grades through deep learning models has become a key problem in grassland degradation monitoring and evaluation. SUMMARY

[0005] To solve the above technical problems, the application provides a grassland degradation monitoring method based on remote sensing monitoring, which can integrate multi-source and multi-temporal remote sensing data, extract dynamic features reflecting grassland vegetation, soil, and human disturbance, and achieve the fusion of spatio-temporal features and the accurate classification of degradation grades through deep learning models.

[0006] To achieve the above purpose, the application provides a grassland degradation monitoring method based on remote sensing monitoring, comprising:

[0007] acquiring multi-source remote sensing data;

[0008] preprocessing the multi-source remote sensing data to obtain preprocessed remote sensing data;

[0009] extract multi-dimensional feature parameters based on the pre-processed remote sensing data, and obtain a multi-dimensional feature parameter set;

[0010] Perform time series analysis on vegetation dynamic characteristics, soil dynamic characteristics, and human disturbance characteristics by using the multi-dimensional feature parameter set in combination with multi-temporal observation data, and construct a dynamic characteristic time series data set;

[0011] According to the dynamic characteristic time series data set, use a convolutional neural network model to deeply fuse spatial features and time series changes in the time series, and generate a spatio-temporal fusion feature set;

[0012] According to the spatio-temporal fusion feature set, obtain a degradation grade division result map;

[0013] Based on the degradation grade division result map, integrate multi-source remote sensing data to obtain a refined grassland degradation evaluation result, and monitor grassland degradation according to the evaluation result.

[0014] Optionally, the multi-source remote sensing data includes optical image data, radar image data, and thermal infrared image data.

[0015] Optionally, the multi-source remote sensing data is pre-processed to obtain pre-processed remote sensing data, including:

[0016] Radiative transfer model is used to perform radiometric correction on the multi-source remote sensing data to obtain a radiometric correction data set;

[0017] A digital elevation model is used to perform terrain analysis on the radiometric correction data set to calculate terrain slope and slope direction, and obtain terrain feature parameters;

[0018] The shadow distribution of complex terrain areas is calculated by using the terrain feature parameters and the solar elevation angle, and the shadow affected area is determined;

[0019] According to the shadow affected area, a terrain adaptability correction data set is generated.

[0020] Optionally, the multi-dimensional feature parameters are extracted based on the pre-processed remote sensing data, and the multi-dimensional feature parameter set is obtained, including:

[0021] Principal component analysis algorithm is used to extract principal components from the pre-processed remote sensing data to reduce dimensionality, and obtain the multi-dimensional feature parameter set.

[0022] Optionally, the time series analysis on vegetation dynamic characteristics, soil dynamic characteristics, and human disturbance characteristics is performed by using the multi-dimensional feature parameter set in combination with multi-temporal observation data, and the dynamic characteristic time series data set is constructed, including:

[0023] Through multi-temporal observation, multi-temporal observation data of vegetation, soil, and human disturbance are obtained;

[0024] The multi-temporal observation data are integrated by using a data fusion technology to generate an integrated data set;

[0025] The integrated data set is processed by using a time series analysis method to generate a change trend of vegetation dynamic characteristics, soil dynamic characteristics and human disturbance characteristics;

[0026] According to the change trend, a sliding window method is used to calculate the dynamic change rate of the vegetation dynamic characteristics and the soil dynamic characteristics, and a dynamic change rate sequence is obtained;

[0027] Based on the dynamic change rate sequence and the change trend of the human disturbance characteristics, a correlation analysis method is used to determine a dynamic characteristic time series data set.

[0028] Optionally, according to the dynamic characteristic time series data set, a convolutional neural network model is used to deeply fuse the spatial features and the time series changes in the time series to generate a spatio-temporal fusion feature set, including:

[0029] According to the dynamic characteristic time series data set, a convolutional neural network model is used to process the dynamic characteristics to extract spatial feature information and determine a spatial distribution pattern;

[0030] According to the spatial distribution pattern, the time series change rule is analyzed, the change trend in the time series is fused, and a time series change feature is obtained;

[0031] If the fluctuation amplitude of the time series change feature exceeds a preset threshold, the change trend is smoothed to obtain stable time series change data;

[0032] The stable time series change data and the spatial distribution pattern are deeply fused to generate a spatio-temporal fusion feature set.

[0033] Optionally, according to the spatio-temporal fusion feature set, the degradation level division result map is obtained, including:

[0034] According to the spatio-temporal fusion feature set, a preliminary degradation area distribution map is obtained;

[0035] Through the preliminary degradation area distribution map, combined with historical data and regional environmental factors, a multi-dimensional verification is performed on the potential degradation area to determine a final degradation area distribution map;

[0036] According to the final degradation area distribution map, a clustering analysis method is used to grade the degradation degree to generate the degradation level division result map.

[0037] Optionally, based on the degradation level division result map, multi-source remote sensing data are integrated to obtain a refined grassland degradation evaluation result, including:

[0038] The spatial interpolation method is used to process the degradation grade division result image to obtain a spatial distribution image.

[0039] Through the spatial distribution image, the spatial variation trend of the degradation grade is extracted, the clustering analysis method is used to identify the degradation hot spot area, and a hot spot area set is determined.

[0040] According to the hot spot area set, time series data is obtained, the variation trend of the degradation grade with time is analyzed, and dynamic change characteristics are obtained.

[0041] According to the dynamic change characteristics, a data fusion method is used to integrate multi-source data and the spatial distribution image to generate a refined evaluation report containing the degradation grade, the hot spot area and the variation trend, and the report content is output.

[0042] Compared with the prior art, the present application has the following advantages and technical effects:

[0043] The present application collects multi-source remote sensing data, performs radiation correction and terrain adaptability correction, and extracts multi-dimensional feature parameters. Time series analysis is performed in combination with multi-temporal observation data to construct a dynamic characteristic time series data set. Convolutional neural network model is used to deeply fuse the space-time characteristics to generate a space-time fusion feature set. According to the feature change trend, the potential degradation area is determined, and multi-dimensional verification is performed in combination with historical data and environmental factors to determine the final degradation area distribution. Clustering analysis is used to grade the degradation degree, and finally a refined grassland degradation evaluation report is output. According to the evaluation result, the grassland degradation is monitored. The present application realizes accurate monitoring of grassland degradation in complex terrain areas, and provides a scientific basis for grassland ecological protection and sustainable utilization. BRIEF DESCRIPTION OF DRAWINGS

[0044] The accompanying drawings, which form a part of this application, are intended to provide further understanding of the application and are incorporated herein in their entirety. The schematic embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0045] Figure 1 is a flow chart of a grassland degradation monitoring method based on remote sensing monitoring according to an embodiment of the present application. DETAILED DESCRIPTION

[0046] It should be noted that the embodiments in the present application and the features in the embodiments can be combined with each other without conflict. The present application will be described in detail below with reference to the drawings and in combination with the embodiments.

[0047] It should be noted that the steps shown in the flowchart of the 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 herein can be executed in an order different from that shown.

[0048] The embodiment proposes a grassland degradation monitoring method based on remote sensing monitoring, as shown in the figure, specifically comprising the following steps: Figure 1

[0049] Obtaining multi-source remote sensing data;

[0050] Preprocessing the multi-source remote sensing data to obtain the preprocessed remote sensing data;

[0051] Extracting multi-dimensional feature parameters based on the preprocessed remote sensing data to obtain a multi-dimensional feature parameter set;

[0052] Through the multi-dimensional feature parameter set, combining multi-temporal observation data, time series analysis is performed on vegetation dynamic characteristics, soil dynamic characteristics and human disturbance characteristics to construct a dynamic characteristic time series data set;

[0053] According to the dynamic characteristic time series data set, a convolutional neural network model is used to deeply fuse the spatial features and time sequence changes in the time series to generate a spatio-temporal fusion feature set;

[0054] According to the spatio-temporal fusion feature set, a degradation grade division result map is obtained;

[0055] Based on the degradation grade division result map, multi-source remote sensing data is integrated to obtain a refined grassland degradation evaluation result, and according to the evaluation result, the grassland degradation is monitored.

[0056] Specifically, the present application collects multi-source remote sensing data, performs radiation correction and terrain adaptability correction, and extracts multi-dimensional feature parameters. Time series analysis is performed in combination with multi-temporal observation data to construct a dynamic characteristic time series data set. The convolutional neural network model is used to deeply fuse the spatio-temporal features to generate a spatio-temporal fusion feature set. According to the feature change trend, the potential degradation area is determined, and the historical data and environmental factors are combined for multi-dimensional verification to determine the final degradation area distribution. The degradation degree is classified by cluster analysis, and finally a refined grassland degradation evaluation report is output, and according to the evaluation result, the grassland degradation is monitored. The present application realizes the accurate monitoring of grassland degradation in complex terrain area, and provides a scientific basis for grassland ecological protection and sustainable utilization.

[0057] Further, the multi-source remote sensing data includes optical image data, radar image data and thermal infrared image data.

[0058] ​Further, the multi-source remote sensing data is preprocessed to obtain preprocessed remote sensing data, including:

[0059] A radiative transfer model is used to perform radiometric correction on the multi-source remote sensing data to obtain a radiometrically corrected data set.

[0060] A digital elevation model is used to perform terrain analysis on the radiometrically corrected data set to calculate terrain slope and aspect, obtaining terrain feature parameters.

[0061] The shadow distribution of complex terrain regions is calculated based on the terrain feature parameters and the solar elevation angle to determine the shadow-affected area.

[0062] According to the shadow-affected area, a terrain adaptability correction data set is generated.

[0063] Specifically, for example, in the processing of multi-source remote sensing data, it is assumed that the original data set of optical images, radar images and thermal infrared images from different satellites is obtained through a data acquisition platform. The optical images may come from a high-resolution satellite with a resolution of 2 meters and a timestamp of May 10, 2023; the radar images come from a synthetic aperture radar satellite with a resolution of 10 meters and the same timestamp; the thermal infrared images have a resolution of 30 meters and a slightly different timestamp. These data are extracted through standardized interface protocols to obtain metadata such as imaging time, sensor parameters, etc., forming a multi-source original data set. Specifically, for radiometric correction, a radiative transfer model is used to process the three types of images. When correcting the optical images, the effects of atmospheric scattering and absorption are considered to convert the digital values of the images to ground reflectance; when correcting the radar images, the effects of terrain undulations on echo intensity are considered; and when correcting the thermal infrared images, temperature radiation values are calibrated to generate a radiometrically corrected data set. This step ensures the consistency of the data in the radiometric layer, laying the foundation for subsequent fusion.

[0064] For example, in the resolution difference processing, it is assumed that the resolution of the optical images is 2 meters, while the resolution of the thermal infrared images is 30 meters. Through a spatial interpolation algorithm such as bilinear interpolation, the low-resolution images are resampled to 2 meters to generate a data set with consistent spatial resolution. This improves the comparability of the data in the spatial scale. Specifically, in the geometric registration step, a feature point-based algorithm is used to spatially align the multi-source images. For example, ground control points are selected, and a transformation matrix is used to correct image shifts to ultimately obtain a geometrically corrected unified data set, ensuring spatial correspondence at the pixel level.

[0065] For example, in data fusion, a principal component analysis algorithm is used to extract color features from optical images, texture features from radar images, and temperature features from thermal infrared images to generate a fusion feature data set. This step effectively integrates multi-source information, highlights key features, and improves data expression capability. Specifically, for optimization of the fusion feature data set, an autoencoder model is used for data compression.

[0066] For example, high-dimensional features are compressed to a preset dimension, such as from 100 dimensions to 20 dimensions, to obtain an optimized feature dataset. This not only reduces storage requirements, but also improves subsequent analysis efficiency.

[0067] For example, if the dimension of the optimized feature dataset meets a threshold, such as being less than 30 dimensions, it is saved to a distributed database through a data storage interface to form a final unified radiation correction dataset. This storage method supports efficient querying and large-scale data management, providing reliable data support for subsequent applications such as disaster monitoring and land use analysis. Through the above process, each technical link is closely linked, ensuring the consistency and availability of data from collection to storage, significantly improving the application value of remote sensing data.

[0068] Further, based on the preprocessed remote sensing data, multi-dimensional feature parameters are extracted to obtain a multi-dimensional feature parameter set, including:

[0069] A principal component analysis algorithm is used to extract principal components from the preprocessed remote sensing data, reducing the dimensionality to obtain a multi-dimensional feature parameter set.

[0070] Further, through the multi-dimensional feature parameter set, combined with multi-temporal observation data, time series analysis is performed on vegetation dynamic characteristics, soil dynamic characteristics, and human disturbance characteristics to construct a dynamic characteristic time series dataset, including:

[0071] Through multi-temporal observation, multi-temporal observation data of vegetation, soil, and human disturbance are obtained;

[0072] Data fusion technology is used to integrate the multi-temporal observation data to generate an integrated dataset;

[0073] The integrated dataset is processed through a time series analysis method to generate the change trend of vegetation dynamic characteristics, soil dynamic characteristics, and human disturbance characteristics;

[0074] According to the change trend, a sliding window method is used to calculate the dynamic change rate of vegetation dynamic characteristics and soil dynamic characteristics to obtain a dynamic change rate sequence;

[0075] Based on the dynamic change rate sequence and the change trend of human disturbance characteristics, a correlation analysis method is used to determine the dynamic characteristic time series dataset.

[0076] Specifically, multi-dimensional feature parameter data of vegetation, soil, and human disturbance are obtained through multi-temporal observation. Remote sensing images combined with ground sensor data are used to collect vegetation coverage, soil moisture content, and human disturbance intensity, such as urban expansion or agricultural activity impact, at different time points. Assuming that in a certain area, Sentinel-2 satellite images in spring and autumn show that the vegetation coverage is 60% and 45%, respectively, and the soil moisture content is 30% and 25%, respectively. The human disturbance intensity is quantified by the amount of night light data, which is 0.4 in spring and 0.6 in autumn. This multi-temporal data forms the original data set, reflecting the dynamic changes in the time dimension.

[0077] In one possible implementation, when integrating multi-temporal data sets, the data fusion technology can use a weighted average method to align data from different sources, such as satellite images and ground station data, according to time stamps. For example, for vegetation coverage, the spectral data of the fused image and the ground measured data are combined to generate a unified multi-dimensional feature parameter set, ensuring data consistency. If there are missing values in the data set, such as missing soil moisture content data at a certain time point, the missing value can be calculated based on the values at the previous and subsequent time points (e.g., 28% and 26%) using linear interpolation, resulting in 27%, thereby completing the data set. Specifically, principal component analysis algorithm can be used to extract vegetation dynamic features, soil dynamic features, and human disturbance features. For example, the analysis results may show that vegetation coverage and normalized difference vegetation index (NDVI) are the principal components, with a contribution rate of 70%, reflecting the dynamic changes of vegetation; soil moisture content and electrical conductivity are the principal components, with a contribution rate of 65%, reflecting the dynamic characteristics of soil; and human disturbance features are extracted through light intensity and land use changes, with a contribution rate of 60%. These features form the feature extraction results after dimension reduction, reducing the computational complexity of subsequent analysis. For example, time series analysis can process the feature extraction results through an autoregressive model to generate a trend. Assuming that the vegetation coverage decreases from 60% to 50% in 6 months, the soil moisture content decreases from 30% to 24%, and the human disturbance intensity increases from 0.4 to 0.7, trend analysis can reveal the seasonal changes of vegetation and soil and the growth trend of human disturbance. When calculating the dynamic change rate using a sliding window method, a 3-month window can be set to calculate the vegetation coverage change rate (e.g., a decrease of 3% per month) and the soil moisture content change rate (e.g., a decrease of 2% per month), generating a dynamic change rate sequence. In one possible implementation, the correlation analysis method can evaluate the influence weight of human disturbance on vegetation and soil dynamic features through the Pearson correlation coefficient. For example, the analysis shows that the correlation coefficient between human disturbance intensity and vegetation coverage change rate is -0.85, and the correlation coefficient between human disturbance intensity and soil moisture content change rate is -0.78, indicating that human disturbance has a significant negative impact on both. The weight can be allocated accordingly, such as human disturbance affecting vegetation with a weight of 0.6 and soil with a weight of 0.4. This analysis helps to quantify the impact of human activities on the ecosystem and provides a basis for subsequent management.

[0078] Further, according to the dynamic feature time series data set, the spatial features and time sequence changes in the time series are deeply fused by using a convolutional neural network model to generate a spatio-temporal fusion feature set, including:

[0079] According to the dynamic feature time series data set, a convolutional neural network model is used to process the dynamic features, extract spatial feature information, and determine the spatial distribution pattern;

[0080] According to the spatial distribution pattern, the time sequence change rule is analyzed, the change trend in the time series is fused, and the time sequence change feature is obtained;

[0081] If the fluctuation amplitude of the time sequence change feature exceeds the preset threshold, the change trend is smoothed to obtain stable time sequence change data;

[0082] The stable time sequence change data and the spatial distribution pattern are deeply fused to generate a spatio-temporal fusion feature set.

[0083] Specifically, by processing the time series data set, dynamic feature information of vegetation coverage, soil moisture, and human activity intensity can be obtained. Assuming that in an agricultural area, satellite remote sensing data is used, vegetation index, soil reflectance, and land use change data are collected every 10 days, the time span is one year, and there are 36 groups of data. The preliminary data structure can be a table containing time stamp, vegetation index, soil moisture, and human activity intensity, constituting a basic data framework. This framework facilitates subsequent analysis and ensures the continuity of data in the time dimension.

[0084] In one possible implementation, a convolutional neural network model is employed to extract spatial feature information. For vegetation coverage and soil moisture data, a multi-channel convolutional neural network is designed, with each channel processing the spatial data of vegetation and soil respectively. For example, the region is divided into 100x100 meter grids, and each grid point contains vegetation index and soil moisture values. The convolutional neural network captures the boundary changes of vegetation coverage or the spatial heterogeneity of soil moisture between grids through convolution kernels, generating spatial distribution patterns. Such patterns can reflect the gradient changes of vegetation coverage from the center of farmland to the edge of the transition zone, or the distribution differences of soil moisture between irrigated and non-irrigated areas. Specifically, when analyzing the temporal variation law, the spatial distribution pattern can be combined with time series data to identify the periodic changes of vegetation coverage or soil moisture. For example, the vegetation index in spring may rise from 0.3 to 0.6, reflecting the crop growth cycle, while soil moisture may increase from 20% to 35% after rainfall. By Fourier transform or autoregressive model, the seasonal variation trend is extracted, and the change characteristics in the time series are fused. If it is found that the vegetation index fluctuation amplitude exceeds the preset threshold of 0.2 in a certain period of time, indicating that it may be affected by extreme weather, then the moving average method is used for smoothing processing to generate stable time series change data. This smoothing processing can reduce noise interference and highlight long-term trends, facilitating subsequent analysis. In one embodiment, the construction of spatio-temporal fusion features can be achieved by combining stable time series change data with spatial distribution patterns. For example, the spatial gradient of vegetation coverage is superimposed with the seasonal variation trend to generate a three-dimensional feature matrix containing time, spatial coordinates, and feature values. This integrated feature representation can fully describe the dynamic variation law of vegetation and soil in the region. When verifying consistency, the correlation coefficient between the integrated feature representation and the original time series data set can be calculated. For example, a correlation coefficient of 0.9 or higher indicates that the fused features effectively retain the information of the original data. The final fused feature set can be used for subsequent monitoring or prediction to provide data support for agricultural management. It should be noted that each step of the above method is closely related to the analysis of vegetation and soil dynamic features, and the generated fused feature set has good spatio-temporal consistency. This consistency helps to accurately identify the dynamic variation law in agricultural areas and provides a basis for optimizing resource allocation. For example, irrigation plans can be adjusted according to the spatial distribution and temporal variation of soil moisture to reduce water waste.

[0085] Further, according to the spatio-temporal fusion feature set, obtaining the degradation level division result map includes:

[0086] According to the spatio-temporal fusion feature set, obtaining a preliminary degradation area distribution map;

[0087] Through the preliminary degradation area distribution map, combined with historical data and regional environmental factors, the potential degradation area is verified in multiple dimensions to determine the final degradation area distribution map;

[0088] According to the final degradation area distribution map, the degradation degree is classified by using a clustering analysis method, and a degradation grade division result map is generated.

[0089] Specifically, by obtaining basic information from the preliminary degradation area distribution map, the distribution map is segmented by using image processing technology, and the degradation area features required for preliminary analysis are extracted, and the preliminary divided area range is obtained.

[0090] According to the preliminary divided area range, the corresponding historical data is obtained, and the historical data is compared with the current area features by using data matching technology to determine whether there is a significant deviation. If the comparison result exceeds the preset threshold, the area range is adjusted, and the corrected area boundary is determined.

[0091] For the corrected area boundary, relevant environmental factor data is obtained, and the environmental factors and area boundary information are integrated by using data fusion technology to obtain a comprehensive influence factor data set.

[0092] According to the comprehensive influence factor data set, a random forest algorithm is used to classify and predict potential areas, to determine which areas have a high risk of degradation, and to determine a potential degradation area list.

[0093] Through the potential degradation area list, the data sources required for multidimensional verification are obtained, and the cross verification of environmental factors and historical data is performed for each potential area. If the verification result meets the preset condition, it is marked as a high-risk area, and a verified area set is obtained.

[0094] According to the verified area set, combined with the area distribution information, the boundary of the high-risk area is optimized by using a spatial analysis tool to generate a final degradation area distribution map.

[0095] More specifically, the data of the final degradation area distribution map is obtained, the feature information of the degradation area is extracted by digitizing the distribution map, and a preliminary degradation feature data set is obtained.

[0096] For the preliminary degradation feature data set, a clustering analysis method is used to classify and process the degradation degree, to determine the category distribution of different degradation degrees, and to form a degradation degree classification result.

[0097] According to the degradation degree classification result, combined with the preset grade standard, each category is classified and processed to determine the grade division of the degradation degree, and a degradation grade data set is obtained.

[0098] If the category boundary in the degradation grade data set is not clear, the degradation features are analyzed again to adjust the classification boundary, and the final grade division result is determined.

[0099] For the final classification results, image generation techniques are used to combine degradation levels and regional classification to generate visual images of degradation level distribution.

[0100] According to the generated visual image, the rationality of regional classification is verified, and if it is found that part of the regional classification does not match the degradation characteristics, the data classification logic is re-adjusted to obtain an optimized result diagram.

[0101] By formatting the optimized result diagram, the image data is converted into a standard output form to determine the final degradation level classification result diagram.

[0102] Further, based on the degradation level classification result diagram, multi-source remote sensing data is integrated to obtain a refined grassland degradation evaluation result, including:

[0103] The spatial distribution image is obtained by using a spatial interpolation method to process the degradation level classification result diagram;

[0104] Through the spatial distribution image, the spatial variation trend of the degradation level is extracted, and a clustering analysis method is used to identify the hot spot area to determine the hot spot area set;

[0105] According to the hot spot area set, time series data is obtained to analyze the change trend of the degradation level with time to obtain dynamic change characteristics;

[0106] According to the dynamic change characteristics, a data fusion method is used to integrate multi-source data and spatial distribution images to generate a refined evaluation report containing degradation levels, hot spot areas and change trends, and the report content is output.

[0107] Specifically, for example, when obtaining remote sensing images, meteorological data and soil parameters from multi-source data, Landsat images can be obtained through satellite remote sensing, combined with rainfall records of meteorological stations and soil humidity sensor data to form an initial data set. The data cleaning method can use outlier detection to remove cloud cover noise in the image or extreme values in the meteorological data. For example, set the rainfall threshold to 0 to 500 mm, and remove the abnormal data outside the range to obtain a standardized data set. This cleaning ensures data consistency and provides a reliable basis for subsequent analysis. In one possible implementation, when extracting dynamic features such as vegetation coverage, soil moisture, and rainfall, vegetation coverage can be calculated by normalized difference vegetation index, assuming that the coverage of a certain area decreases from 0.6 to 0.3, reflecting the degradation trend. Soil moisture can be obtained through sensor data, such as a point with a humidity of 20% and an annual rainfall of 300 mm.

[0108] When using principal component analysis for dimension reduction, these features can be converted into 2 to 3 principal components, retaining more than 90% of the data variance, simplifying the analysis while highlighting key features. For example, for the key feature set, if the vegetation coverage is below the preset threshold of 0.4, combined with the conditions of soil moisture below 25% and rainfall less than 200 mm, the support vector machine algorithm can be used to classify the degradation level.

[0109] Assuming that a certain area meets these conditions, the algorithm classifies it as moderately degraded, generating a distribution result containing light, moderate, and severe degradation. This method clearly classifies the levels through the classification boundary, facilitating subsequent spatial analysis.

[0110] In one possible implementation, the spatial interpolation method can use Kriging interpolation to generate a continuous degradation level distribution map based on existing degradation level data points. For example, the central part of a certain area shows severe degradation, and the edge is lightly degraded, forming a smooth spatial distribution image after interpolation. This image intuitively shows the spatial variation of the degradation level, facilitating the identification of key areas. For example, when extracting the spatial variation trend of the degradation level, the K-means algorithm can be used to identify hotspots by analyzing the trend of degradation decreasing from the center to the edge.

[0111] For example, three hotspots in a certain area show severe degradation, with a coverage rate of less than 0.2 and soil moisture of less than 15%. These hotspot regions provide targets for subsequent dynamic analysis. In one possible implementation, when obtaining time series data of the hotspot regions, the change in vegetation coverage from 0.5 to 0.3 over the past 5 years can be analyzed, combined with the trend of rainfall decreasing from 400 mm to 200 mm, to obtain the dynamic characteristics of the intensifying degradation. This analysis reveals the variation of degradation over time, providing a basis for long-term monitoring. For example, the data fusion method can integrate remote sensing images, hotspots, and trends to generate a refined evaluation report. The report can show that the central part of a certain area is severely degraded, with a coverage rate of less than 0.2, combined with the trend of decreasing rainfall, and propose key management recommendations. This report integrates multi-source information, comprehensively presents the degradation status and dynamic changes, and provides support for decision-making.

[0112] The above is only a preferred specific implementation of the present application, but the protection scope of the present application is not limited thereto, and any person skilled in the art can easily think of changes or replacements within the technical scope disclosed in the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A method for monitoring grassland degradation based on remote sensing, characterized in that, include: Acquire multi-source remote sensing data; The multi-source remote sensing data is preprocessed to obtain preprocessed remote sensing data; Based on the preprocessed remote sensing data, multidimensional feature parameters are extracted to obtain a multidimensional feature parameter set. By combining multi-dimensional feature parameter sets with multi-temporal observation data, time series analysis is performed on vegetation dynamics, soil dynamics, and human disturbance characteristics to construct a dynamic feature time series dataset. By combining a multidimensional feature parameter set with multi-temporal observation data, time series analysis is performed on vegetation dynamics, soil dynamics, and anthropogenic disturbance characteristics, constructing a dynamic feature time series dataset including: Multi-temporal observations were conducted to obtain multi-temporal observation data on vegetation, soil, and human disturbance. The multi-temporal observation data are integrated using data fusion technology to generate an integrated dataset; The integrated dataset was processed using time series analysis methods to generate trends in vegetation dynamics, soil dynamics, and anthropogenic disturbance characteristics. Based on the aforementioned trend, the dynamic change rate of vegetation dynamic characteristics and soil dynamic characteristics is calculated using the sliding window method to obtain a dynamic change rate sequence. Based on the changing trends of the dynamic rate of change sequence and the characteristics of human interference, a correlation analysis method is used to determine the dynamic characteristic time series dataset; Based on the dynamic feature time series dataset, a convolutional neural network model is used to deeply fuse the spatial features and temporal changes in the time series to generate a spatiotemporal fusion feature set. Based on the dynamic feature time series dataset, a convolutional neural network model is used to deeply fuse the spatial features and temporal changes in the time series, generating a spatiotemporal fusion feature set including: Based on the dynamic feature time series dataset, a convolutional neural network model is used to process the dynamic features, extract spatial feature information, and determine the spatial distribution pattern. Based on the spatial distribution pattern, analyze the temporal variation law, integrate the variation trend in the time series, and obtain the temporal variation characteristics; If the fluctuation amplitude of the time series change characteristics exceeds the preset threshold, the change trend is smoothed to obtain stable time series change data; By deeply fusing stable temporal variation data with spatial distribution patterns, a spatiotemporal fusion feature set is generated. Based on the spatiotemporal fusion feature set, obtain the degradation level classification result map; Based on the spatiotemporal fusion feature set, the degradation level classification result map is obtained as follows: Based on the spatiotemporal fusion feature set, a preliminary degradation region distribution map is obtained; By combining the preliminary degradation area distribution map with historical data and regional environmental factors, potential degradation areas are verified in multiple dimensions to determine the final degradation area distribution map. Based on the final degradation area distribution map, cluster analysis is used to classify the degree of degradation, generating the degradation level classification result map; Based on the degradation level classification map, multi-source remote sensing data are integrated to obtain refined grassland degradation assessment results, and grassland degradation is monitored according to the assessment results.

2. The grassland degradation monitoring method based on remote sensing according to claim 1, characterized in that, The multi-source remote sensing data includes: optical image data, radar image data, and thermal infrared image data.

3. The grassland degradation monitoring method based on remote sensing according to claim 1, characterized in that, Preprocessing the multi-source remote sensing data to obtain preprocessed remote sensing data includes: The multi-source remote sensing data were radiometrically corrected using a radiative transfer model to obtain a radiometrically corrected dataset. A digital elevation model was used to perform terrain analysis on the radiometrically corrected dataset, calculate the terrain slope and aspect, and obtain terrain feature parameters. The shadow distribution in complex terrain areas is calculated using the terrain feature parameters and solar altitude angle to determine the shadow-affected areas; A terrain-adaptive correction dataset is generated based on the shadow-affected area.

4. The grassland degradation monitoring method based on remote sensing according to claim 1, characterized in that, Based on the preprocessed remote sensing data, multidimensional feature parameters are extracted to obtain a multidimensional feature parameter set, including: Principal component analysis (PCA) is used to extract principal components from the preprocessed remote sensing data, reduce dimensionality, and obtain the multidimensional feature parameter set.

5. The grassland degradation monitoring method based on remote sensing according to claim 1, characterized in that, Based on the degradation level classification map, multi-source remote sensing data are integrated to obtain refined grassland degradation assessment results, including: The degradation level classification result map is processed using a spatial interpolation method to obtain a spatial distribution image; Using the spatial distribution image, the spatial variation trend of degradation level is extracted, and cluster analysis is used to identify degradation hotspot areas and determine the set of hotspot areas; Based on the set of hotspot areas, time series data are obtained, and the trend of degradation level changes over time is analyzed to obtain dynamic change characteristics. Based on the dynamic change characteristics, a data fusion method is used to integrate multi-source data and spatial distribution images to generate a refined assessment report containing degradation level, hotspot areas and change trends, and output the report content.

Images