Remote sensing identification method and system for abandoned land of slope cropland

Abstract

The invention discloses a remote sensing recognition method and system for abandoned sloping farmland, and the system comprises a data preprocessing module which is responsible for collecting and processing multi-source remote sensing image data, including Landsat and Sentinel-2 satellite data; the deep learning model training module is used for carrying out classification training on the remote sensing images by utilizing a UNet + + network and other deep learning algorithms; and the comprehensive interpretation model construction module is used for realizing comprehensive interpretation of the abandoned land of the slope cropland in combination with a typical terrain accurate classification and fuzzy classification technology. Through a deep learning algorithm and a multi-source data fusion technology, the system realizes high-precision abandoned slope cropland recognition, and the detection efficiency and accuracy are improved; the system integrates various remote sensing data and machine learning algorithms, and can comprehensively and accurately identify different types of land cover changes, including spring crops, summer crops and the like.

Description

Technical Field

[0001] The present invention relates to the field of rocker technology, and in particular to a remote sensing identification method and system for abandoned sloping farmland. Background Art

[0002] The monitoring and identification of sloping farmland abandonment is an important topic in current agricultural land management and food security research. With the advancement of urbanization and the reduction of rural labor force, a large amount of sloping farmland has been gradually abandoned, resulting in the waste of farmland resources and the decline of food production capacity. Therefore, it is particularly important to develop an efficient and accurate remote sensing identification method to monitor and evaluate the sloping farmland abandonment phenomenon.

[0003] To this end, the present invention proposes a remote sensing identification method and system for abandoned sloping farmland. Summary of the invention

[0004] In view of the defects in the prior art, the present invention provides a remote sensing identification system for abandoned sloping farmland, comprising:

[0005] Data preprocessing module: responsible for collecting and processing multi-source remote sensing image data, including Landsat and Sentinel-2 satellite data;

[0006] Deep learning model training module: Use UNet++ network and other deep learning algorithms to classify remote sensing images;

[0007] Comprehensive interpretation model building module: Combines typical ground object accurate classification and fuzzy classification technology to achieve comprehensive interpretation of sloping farmland abandonment;

[0008] Spatiotemporal distribution map drawing module: based on the trained model, draw the spatiotemporal distribution map of abandoned sloping farmland;

[0009] Risk probability distribution map drawing module: using XGBoost random forest algorithm, based on the influencing factors at the plot scale, draw the risk probability distribution map of sloping farmland abandonment;

[0010] Food production capacity impact assessment module: Combined with arable land quality data and major grain crop yield data, the module estimates the impact of sloping farmland abandonment on food production capacity.

[0011] A remote sensing identification method for abandoned sloping farmland is implemented based on the above remote sensing identification system. The remote sensing identification method is based on the data preprocessing module of the remote sensing identification system. When performing data processing, in the data preprocessing stage, multi-source remote sensing image data of Landsat 7ETM+, Landsat 8OLI, Sentinel-2, and MODIS are first obtained from the Google Earth Engine platform; after the image is atmospherically corrected, the normalized vegetation index and other related vegetation indices are used to extract spectral features; for areas with severe cloud and fog obstruction, a multi-source data fusion method is used to fill in the gaps, and an SG filter is used for correction;

[0012] The remote sensing recognition method is based on a deep learning model training module, uses the UNet++ network as the main deep learning framework, and uses high-definition remote sensing images as the data source for model training; the specific steps are as follows:

[0013] a) Sample selection and annotation: According to the geomorphological characteristics of the study area, representative sample areas are selected for manual annotation;

[0014] b) Data preprocessing: standardize the labeled sample data;

[0015] c) Model training: Use the Adam optimizer and gradient descent strategy to train the UNet++ model;

[0016] d) Accuracy verification: The accuracy of the trained model is verified through confusion matrix and Kappa coefficient.

[0017] Preferably: the remote sensing identification method is based on a comprehensive interpretation model construction module. In the comprehensive interpretation model construction stage, the strategy of accurate classification of typical objects and fuzzy classification of machine learning algorithms is combined to conduct a comprehensive interpretation of abandoned sloping farmland; the details are as follows:

[0018] a) Identification of spring crops: Use object-oriented + decision tree method to identify the growth characteristics of spring crops;

[0019] b) Extraction of the multiple cropping index: The multiple cropping index is calculated by smoothing the seamless NDVI time series;

[0020] c) Extraction of rice terraces: Identify rice terraces based on the characteristics of the rice growth cycle;

[0021] d) Extraction of summer crops: Random forest + long time series vegetation features are used to identify summer crops;

[0022] e) Comprehensive interpretation of land cover types: Superimpose and analyze the land cover type maps of different periods to determine whether there is any phenomenon of abandoned sloping farmland;

[0023] The remote sensing identification method, based on the spatiotemporal distribution map drawing module, draws the spatiotemporal distribution map of abandoned terraces according to the trained model; the specific steps are as follows:

[0024] a) Initial distribution map generation: using the initially extracted terrace distribution map as a basis, combined with the results of the comprehensive interpretation model for analysis;

[0025] b) Error analysis and correction: Conduct error analysis on the preliminary results and make corrections based on field survey data;

[0026] c) Final distribution map generation: Generate a high-precision spatiotemporal distribution map.

[0027] Preferably, the remote sensing identification method, based on the risk probability distribution map drawing module, uses the XGBoost random forest algorithm to draw a risk probability distribution map of abandoned agricultural terraces according to the influencing factors at the plot scale; the specific steps are as follows:

[0028] a) Statistics of influencing factors: Data statistics of terraced fields across the country were collected, including the distance to the nearest rural settlement, the distance to the nearest road, the distance to the water source, and the difference in cultivation height;

[0029] b) Model training and verification: On the GEE platform, the XGBoost algorithm was used to conduct a probability assessment of terraced fields that had not been abandoned, and the results were verified;

[0030] c) Risk distribution map generation: The assessment results are combined with spatial area weights to generate a spatial grid distribution map to intuitively display the risk level of each area.

[0031] Preferably: the remote sensing identification method, based on the food production capacity impact assessment module, combines the cultivated land quality distribution map and the yield data of major food crops to estimate the impact of the current agricultural terrace abandonment on China's food production capacity; the specific steps are as follows:

[0032] a) Calculation of grain loss: The grain loss caused by abandonment of farmland in each province is calculated using the formula Qp = Sap × a × Yieldp;

[0033] b) Regional food demand comparison: compare the reduction amount with the basic food demand of the region to assess the actual impact of abandonment;

[0034] c) Estimation of potential impacts: Set different abandonment scenarios under different scenarios to estimate the potential amount of food loss in the future and its impact on food security.

[0035] Preferably, the remote sensing recognition method uses the Adam optimizer and the gradient descent strategy to train the UNet++ model, specifically comprising the following steps:

[0036] S1: Data collection and preprocessing: First, collect a large amount of high-definition remote sensing image data, and perform atmospheric correction and cloud removal data preprocessing operations; divide the image data into training set and validation set;

[0037] S2: Model construction: UNet++ network is used as the main deep learning framework;

[0038] S3: Parameter setting: The learning rate is initially set to 0.001, and converges quickly to the target area within the first 500 generations; then the learning rate is reduced to 0.0001, and fine-tuned to improve the classification accuracy; the Adam optimizer is used as the gradient descent strategy, and momentum and adaptive learning rate are combined to accelerate the convergence speed. The cross entropy classification loss function is used as the objective function, and the loss values ​​of terraces and non-terraces are multiplied by 5 and 1 respectively;

[0039] S4: Training and evaluation: Use labeled training samples to train the model, and continuously monitor the training progress and loss changes to ensure that the model achieves optimal performance within the scheduled time;

[0040] S5: Accuracy verification and correction: The model is verified for accuracy through confusion matrix and Kappa coefficient. Based on the verification results, the model is fine-tuned or regularization is added to prevent overfitting.

[0041] Preferably, the remote sensing identification method uses object-oriented + decision tree method to identify the growth characteristics of spring crops, and the specific steps are:

[0042] S1: Spectral feature extraction: extract spectral features from the smoothed seamless NDVI time series after SG filtering;

[0043] S2: Object-oriented segmentation: Use object-oriented image processing methods to segment the image into multiple independent object regions;

[0044] S3: Application of decision tree method: Based on the spectral characteristics of each object, a decision tree model is constructed to determine whether the object is a spring crop by setting a specific threshold;

[0045] S4: Integration of results: Comparative analysis of identified spring crops with other types of land cover.

[0046] Preferably, the remote sensing identification method calculates the multiple cropping index by smoothing and seamless NDVI time series, and the specific steps are:

[0047] S1: Data preparation: Collect historical Landsat or Sentinel-2 image data to generate seamless time series NDVI data;

[0048] S2: Smoothing: Use smoothing seamless technology to remove noise in the time series and generate a smooth NDVI curve;

[0049] S3: Calculation of the multiple cropping index: Based on the smoothed NDVI time series, the multiple cropping index within a specific time period is calculated as follows:

[0050]

[0051] Where t is the study duration and FPAR(t') is the FPAR value at time t'.

[0052] Preferably, the remote sensing identification method adopts random forest + long time series vegetation feature mode to identify summer crops, and the specific steps are as follows:

[0053] S1: Data preparation: Collect multi-year Landsat or Sentinel-2 image data to generate long-term NDVI or other vegetation index datasets;

[0054] S2: Multi-source data fusion: NDVI data from different periods are fused together to supplement the data missing due to cloud cover;

[0055] S3: Random forest modeling: Use the random forest algorithm to model the long-term vegetation index data and establish a classification model to distinguish summer crops from other landforms;

[0056] S4: Model training and verification: Model training and verification are performed on the GEE platform to ensure that the model has good generalization capabilities.

[0057] Preferably, the remote sensing identification method sets different abandonment scenarios under different scenarios to estimate the potential future food loss and its impact on food security, and the specific steps are as follows:

[0058] S1: Scenario setting: Different abandonment scenarios are set based on historical data analysis and policy assumptions, such as complete abandonment, partial abandonment and no abandonment;

[0059] S2: Calculation of food loss: The food loss under each scenario is calculated using the formula Qp = Sap × a × Yieldp, where Qp is the loss, Sap is the affected cultivated land area, a is the yield reduction ratio, and Yieldp is the original yield;

[0060] S3: Risk assessment: Comprehensively assess food security risks under various scenarios by combining regional population growth and economic development factors;

[0061] S4: Policy recommendation formulation: Propose corresponding agricultural management and land protection policy recommendations based on the results of scenario analysis.

[0062] The beneficial effects of the present invention are embodied in:

[0063] 1. Through deep learning algorithms and multi-source data fusion technology, the system achieves high-precision identification of abandoned sloping farmland and improves detection efficiency and accuracy.

[0064] 2. The system of the present invention integrates a variety of remote sensing data and machine learning algorithms, and can comprehensively and accurately identify different types of land cover changes, including spring crops, summer crops, etc. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following is a brief introduction to the drawings required for the specific embodiments or the description of the prior art. In all the drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, the elements or parts are not necessarily drawn according to the actual scale.

[0066] Figure 1 This is a flow chart of the present invention using high-definition remote sensing images as a data source for model training;

[0067] Figure 2 A flow chart of the comprehensive interpretation of sloping farmland abandonment performed by the present invention;

[0068] Figure 3 A flow chart for drawing a spatiotemporal distribution map of abandoned terraced fields for the present invention;

[0069] Figure 4 A flow chart for drawing a risk probability distribution map of abandoned agricultural terraces for the present invention;

[0070] Figure 5 The present invention is a flow chart for estimating the impact of current agricultural terrace abandonment on China's grain production capacity. DETAILED DESCRIPTION

[0071] The following embodiments of the technical solution of the present invention are described in detail in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and are therefore only used as examples, and cannot be used to limit the protection scope of the present invention.

[0072] It should be noted that, unless otherwise specified, the technical terms or scientific terms used in this application should have the common meanings understood by those skilled in the art to which the present invention belongs.

[0073] Embodiment 1:

[0074] A remote sensing identification method and system for abandoned sloping farmland, the system comprising:

[0075] Data preprocessing module: responsible for collecting and processing multi-source remote sensing image data, including Landsat, Sentinel-2 and other satellite data;

[0076] Deep learning model training module: Use UNet++ network and other deep learning algorithms to classify remote sensing images;

[0077] Comprehensive interpretation model building module: Combines typical ground object accurate classification and fuzzy classification technology to achieve comprehensive interpretation of sloping farmland abandonment;

[0078] Spatiotemporal distribution map drawing module: based on the trained model, draw the spatiotemporal distribution map of abandoned sloping farmland;

[0079] Risk probability distribution map drawing module: using XGBoost random forest algorithm, based on the influencing factors at the plot scale, draw the risk probability distribution map of sloping farmland abandonment;

[0080] Food production capacity impact assessment module: Combined with arable land quality data and major grain crop yield data, the module estimates the impact of sloping farmland abandonment on food production capacity.

[0081] The remote sensing identification method is based on a data preprocessing module of a remote sensing identification system. When performing data processing, in the data preprocessing stage, multi-source remote sensing image data such as Landsat 7ETM+, Landsat 8OLI, Sentinel-2, MODIS, etc. are first obtained from the Google Earth Engine platform; after the image is atmospherically corrected, the normalized difference vegetation index (NDVI) and other related vegetation indices are used to extract spectral features; for areas with severe cloud and fog obstruction, a multi-source data fusion method is used to fill in the gaps, and an SG filter is used for correction.

[0082] The remote sensing recognition method is based on a deep learning model training module, uses the UNet++ network as the main deep learning framework, and uses high-definition remote sensing images as the data source for model training; the specific steps are as follows:

[0083] a) Sample selection and annotation: According to the geomorphic characteristics of the study area, select representative sample areas for manual annotation; ensure that the sample areas contain rich texture information and terrain changes;

[0084] b) Data preprocessing: standardize the labeled sample data to enhance the training effect of the model;

[0085] c) Model training: The UNet++ model is trained using the Adam optimizer and gradient descent strategy to improve classification accuracy and convergence speed;

[0086] d) Accuracy verification: The trained model is verified for accuracy through confusion matrix and Kappa coefficient to ensure that the model has good generalization ability.

[0087] The remote sensing identification method is based on a comprehensive interpretation model construction module. In the comprehensive interpretation model construction stage, the strategy of accurate classification of typical objects and fuzzy classification of machine learning algorithms is combined to conduct a comprehensive interpretation of abandoned sloping farmland. Specifically, it includes the following aspects:

[0088] a) Identification of spring crops: Use object-oriented + decision tree method to identify the growth characteristics of spring crops;

[0089] b) Extraction of the multiple cropping index: The multiple cropping index is calculated by smoothing the seamless NDVI time series;

[0090] c) Extraction of rice terraces: Identify rice terraces based on the characteristics of the rice growth cycle;

[0091] d) Extraction of summer crops: Random forest + long time series vegetation features are used to identify summer crops;

[0092] e) Comprehensive interpretation of land cover types: Superimpose and analyze the land cover type maps of different periods to determine whether there is any abandonment of sloping farmland.

[0093] The remote sensing identification method, based on the spatiotemporal distribution map drawing module, draws the spatiotemporal distribution map of abandoned terraces according to the trained model; the specific steps are as follows:

[0094] a) Initial distribution map generation: using the initially extracted terrace distribution map as a basis, combined with the results of the comprehensive interpretation model for analysis;

[0095] b) Error analysis and correction: Conduct error analysis on the preliminary results and make corrections based on field survey data;

[0096] c) Final distribution map generation: Generate high-precision spatiotemporal distribution maps to provide basic data for further analysis and application.

[0097] The remote sensing identification method, based on the risk probability distribution map drawing module, uses the XGBoost random forest algorithm to draw a risk probability distribution map of abandoned agricultural terraces according to the influencing factors at the plot scale; the specific steps are as follows:

[0098] a) Statistics of influencing factors: Data statistics of terraced fields across the country were collected, including factors such as the distance to the nearest rural settlement, the distance to the nearest road, the distance to the water source, and the height difference of the cultivated land;

[0099] b) Model training and verification: On the GEE platform, the XGBoost algorithm was used to conduct a probability assessment of terraced fields that had not been abandoned, and the results were verified;

[0100] c) Risk distribution map generation: The assessment results are combined with spatial area weights to generate a 10km×10km spatial grid distribution map to intuitively display the risk level of each area.

[0101] The remote sensing identification method, based on the food production capacity impact assessment module, combined with the cultivated land quality distribution map and the yield data of major food crops, estimates the impact of the current agricultural terrace abandonment on China's food production capacity; the specific steps are as follows:

[0102] a) Calculation of grain loss: The grain loss caused by abandonment of farmland in each province is calculated using the formula Qp = Sap × a × Yieldp;

[0103] b) Regional food demand comparison: compare the reduction amount with the basic food demand of the region to assess the actual impact of abandonment;

[0104] c) Estimation of potential impacts: Set different abandonment scenarios under different scenarios to estimate the potential amount of food loss in the future and its impact on food security.

[0105] The remote sensing recognition method uses the Adam optimizer and the gradient descent strategy to train the UNet++ model, and specifically includes the following steps:

[0106] S1: Data collection and preprocessing: First, collect a large amount of high-definition remote sensing image data, and perform data preprocessing operations such as atmospheric correction and cloud removal; divide the image data into training set and validation set;

[0107] S2: Model construction: UNet++ network is used as the main deep learning framework, the input image size is set to 400×400 pixels, and the batch size is set to 6 to balance the model training speed and device memory requirements;

[0108] S3: Parameter setting: The learning rate is initially set to 0.001, and converges quickly to the target area within the first 500 generations; then the learning rate is reduced to 0.0001, and fine-tuned to improve the classification accuracy; the Adam optimizer is used as the gradient descent strategy, and momentum and adaptive learning rate are combined to accelerate the convergence speed. The cross entropy classification loss function is used as the objective function, and the loss values ​​of terraces and non-terraces are multiplied by 5 and 1 respectively;

[0109] S4: Training and evaluation: Use labeled training samples to train the model, and continuously monitor the training progress and loss changes to ensure that the model achieves optimal performance within the scheduled time;

[0110] S5: Accuracy verification and correction: The model is verified for accuracy through confusion matrix and Kappa coefficient. Based on the verification results, the model is fine-tuned or regularization is added to prevent overfitting.

[0111] The remote sensing identification method uses object-oriented + decision tree method to identify the growth characteristics of spring crops. The specific steps are as follows:

[0112] S1: Spectral feature extraction: extract spectral features from the smoothed seamless NDVI time series after SG filtering;

[0113] S2: Object-oriented segmentation: Use object-oriented image processing methods to segment the image into multiple independent object regions;

[0114] S3: Application of decision tree method: Based on the spectral characteristics of each object, a decision tree model is constructed to determine whether the object is a spring crop by setting a specific threshold;

[0115] S4: Result integration: Compare and analyze the identified spring crops with other types of land cover to ensure the accuracy of the final results.

[0116] The remote sensing identification method calculates the multiple cropping index by smoothing the seamless NDVI time series, and the specific steps are as follows:

[0117] S1: Data preparation: Collect historical Landsat or Sentinel-2 image data to generate seamless time series NDVI data;

[0118] S2: Smoothing: Use smoothing techniques (such as SG filter) to remove noise in the time series and generate a smooth NDVI curve;

[0119] S3: Calculation of the multiple cropping index: Based on the smoothed NDVI time series, the cumulative photosynthetically active radiation (FPAR) in a specific time period, namely the multiple cropping index (CI), is calculated as follows:

[0120]

[0121] Where t is the study duration and FPAR(t') is the FPAR value at time t'.

[0122] The remote sensing identification method adopts random forest + long time series vegetation feature method to identify summer crops, and the specific steps are as follows:

[0123] S1: Data preparation: Collect multi-year Landsat or Sentinel-2 image data to generate long-term NDVI or other vegetation index datasets;

[0124] S2: Multi-source data fusion: NDVI data from different periods are fused together to supplement the data missing due to cloud cover;

[0125] S3: Random forest modeling: Use the random forest algorithm to model the long-term vegetation index data and establish a classification model to distinguish summer crops from other landforms;

[0126] S4: Model training and verification: Model training and verification are performed on the GEE platform to ensure that the model has good generalization capabilities.

[0127] The remote sensing identification method sets different abandonment scenarios under different scenarios to estimate the potential future food loss and its impact on food security. The specific steps are as follows:

[0128] S1: Scenario setting: Different abandonment scenarios are set based on historical data analysis and policy assumptions, such as complete abandonment, partial abandonment and no abandonment;

[0129] S2: Calculation of food loss: The food loss under each scenario is calculated using the formula Qp = Sap × a × Yieldp, where Qp is the loss, Sap is the affected cultivated land area, a is the yield reduction ratio, and Yieldp is the original yield;

[0130] S3: Risk assessment: Comprehensively assess food security risks under various scenarios by combining regional population growth, economic development and other factors;

[0131] S4: Policy recommendation formulation: Propose corresponding agricultural management and land protection policy recommendations based on the results of scenario analysis.

[0132] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein by equivalents. These modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be included in the scope of the claims and specification of the present invention.

Claims

1. A remote sensing identification system for abandoned sloping farmland, characterized in that: The system comprises: Data preprocessing module: responsible for collecting and processing multi-source remote sensing image data, including Landsat and Sentinel-2 satellite data; Deep learning model training module: Use UNet++ network and other deep learning algorithms to classify remote sensing images; Comprehensive interpretation model building module: Combines typical ground object accurate classification and fuzzy classification technology to achieve comprehensive interpretation of sloping farmland abandonment; Spatiotemporal distribution map drawing module: based on the trained model, draw the spatiotemporal distribution map of abandoned sloping farmland; Risk probability distribution map drawing module: using XGBoost random forest algorithm, based on the influencing factors at the plot scale, draw the risk probability distribution map of sloping farmland abandonment; Food production capacity impact assessment module: Combined with arable land quality data and major grain crop yield data, the module estimates the impact of sloping farmland abandonment on food production capacity.

2. A remote sensing identification method for abandoned sloping farmland, characterized in that: According to the remote sensing identification system of claim 1, the remote sensing identification method is based on the data preprocessing module of the remote sensing identification system. When performing data processing, in the data preprocessing stage, multi-source remote sensing image data of Landsat 7ETM+, Landsat 8OLI, Sentinel-2, and MODIS are first obtained from the Google Earth Engine platform; after the image is atmospherically corrected, the normalized vegetation index and other related vegetation indices are used to extract spectral features; For areas with severe cloud and fog obstruction, multi-source data fusion method is used to fill in the gaps, and SG filter is used for correction; The remote sensing recognition method is based on a deep learning model training module, uses the UNet++ network as the main deep learning framework, and uses high-definition remote sensing images as the data source for model training; The specific steps are as follows: a) Sample selection and annotation: According to the geomorphological characteristics of the study area, representative sample areas are selected for manual annotation; b) Data preprocessing: standardize the labeled sample data; c) Model training: Use the Adam optimizer and gradient descent strategy to train the UNet++ model; d) Accuracy verification: The accuracy of the trained model is verified through confusion matrix and Kappa coefficient.

3. The remote sensing identification method and system for abandoned sloping farmland according to claim 2, characterized in that: The remote sensing identification method is based on a comprehensive interpretation model construction module. In the comprehensive interpretation model construction stage, the strategy of accurate classification of typical objects and fuzzy classification of machine learning algorithms is combined to conduct a comprehensive interpretation of abandoned sloping farmland; the details are as follows: a) Identification of spring crops: Use object-oriented + decision tree method to identify the growth characteristics of spring crops; b) Extraction of the multiple cropping index: The multiple cropping index is calculated by smoothing the seamless NDVI time series; c) Extraction of rice terraces: Identify rice terraces based on the characteristics of the rice growth cycle; d) Extraction of summer crops: Random forest + long time series vegetation features are used to identify summer crops; e) Comprehensive interpretation of land cover types: Superimpose and analyze the land cover type maps of different periods to determine whether there is any phenomenon of abandoned sloping farmland; The remote sensing identification method, based on the spatiotemporal distribution map drawing module, draws the spatiotemporal distribution map of abandoned terraces according to the trained model; the specific steps are as follows: a) Initial distribution map generation: using the initially extracted terrace distribution map as a basis, combined with the results of the comprehensive interpretation model for analysis; b) Error analysis and correction: Conduct error analysis on the preliminary results and make corrections based on field survey data; c) Final distribution map generation: Generate a high-precision spatiotemporal distribution map.

4. The remote sensing identification method and system for abandoned sloping farmland according to claim 3, characterized in that: The remote sensing identification method, based on the risk probability distribution map drawing module, uses the XGBoost random forest algorithm to draw a risk probability distribution map of abandoned agricultural terraces according to the influencing factors at the plot scale; the specific steps are as follows: a) Statistics of influencing factors: Data statistics of terraced fields across the country were collected, including the distance to the nearest rural settlement, the distance to the nearest road, the distance to the water source, and the difference in cultivation height; b) Model training and verification: On the GEE platform, the XGBoost algorithm was used to conduct a probability assessment of terraced fields that had not been abandoned, and the results were verified; c) Risk distribution map generation: The assessment results are combined with spatial area weights to generate a spatial grid distribution map to intuitively display the risk level of each area.

5. The remote sensing identification method and system for abandoned sloping farmland according to claim 4, characterized in that: The remote sensing identification method, based on the food production capacity impact assessment module, combined with the cultivated land quality distribution map and the yield data of major food crops, estimates the impact of the current agricultural terrace abandonment on China's food production capacity; the specific steps are as follows: a) Calculation of grain loss: The grain loss caused by abandonment of farmland in each province is calculated using the formula Qp = Sap × a × Yieldp; b) Regional food demand comparison: compare the reduction amount with the basic food demand of the region to assess the actual impact of abandonment; c) Estimation of potential impacts: Set different abandonment scenarios under different scenarios to estimate the potential amount of food loss in the future and its impact on food security.

6. The remote sensing identification method and system for abandoned sloping farmland according to claim 5, characterized in that: The remote sensing recognition method uses the Adam optimizer and the gradient descent strategy to train the UNet++ model, and specifically includes the following steps: S1: Data collection and preprocessing: First, collect a large amount of high-definition remote sensing image data, and perform atmospheric correction and cloud removal data preprocessing operations; divide the image data into training set and validation set; S2: Model construction: UNet++ network is used as the main deep learning framework; S3: Parameter setting: The learning rate is initially set to 0.001, and converges quickly to the target region within the first 500 generations; then the learning rate is reduced to 0.0001 and fine-tuned to improve classification accuracy; The Adam optimizer is used as the gradient descent strategy, and momentum and adaptive learning rate are combined to accelerate the convergence speed. The cross entropy classification loss function is used as the objective function, and the loss values ​​of terraces and non-terraces are multiplied by 5 and 1 respectively; S4: Training and evaluation: Use labeled training samples to train the model, and continuously monitor the training progress and loss changes to ensure that the model achieves optimal performance within the scheduled time; S5: Accuracy verification and correction: The model is verified for accuracy through confusion matrix and Kappa coefficient. Based on the verification results, the model is fine-tuned or regularization is added to prevent overfitting.

7. The remote sensing identification method and system for abandoned sloping farmland according to claim 3, characterized in that: The remote sensing identification method uses object-oriented + decision tree method to identify the growth characteristics of spring crops. The specific steps are as follows: S1: Spectral feature extraction: extract spectral features from the smoothed seamless NDVI time series after SG filtering; S2: Object-oriented segmentation: Use object-oriented image processing methods to segment the image into multiple independent object regions; S3: Application of decision tree method: Based on the spectral characteristics of each object, a decision tree model is constructed to determine whether the object is a spring crop by setting a specific threshold; S4: Integration of results: Comparative analysis of identified spring crops with other types of land cover.

8. The remote sensing identification method and system for abandoned sloping farmland according to claim 3, characterized in that: The remote sensing identification method calculates the multiple cropping index by smoothing the seamless NDVI time series, and the specific steps are as follows: S1: Data preparation: Collect historical Landsat or Sentinel-2 image data to generate seamless time series NDVI data; S2: Smoothing: Use smoothing seamless technology to remove noise in the time series and generate a smooth NDVI curve; S3: Calculation of the multiple cropping index: Based on the smoothed NDVI time series, the multiple cropping index within a specific time period is calculated as follows: Where t is the study duration and FPAR(t') is the FPAR value at time t'.

9. The remote sensing identification method and system for abandoned sloping farmland according to claim 3, characterized in that: The remote sensing identification method adopts random forest + long time series vegetation feature method to identify summer crops, and the specific steps are as follows: S1: Data preparation: Collect multi-year Landsat or Sentinel-2 image data to generate long-term NDVI or other vegetation index datasets; S2: Multi-source data fusion: NDVI data from different periods are fused together to supplement the data missing due to cloud cover; S3: Random forest modeling: Use the random forest algorithm to model the long-term vegetation index data and establish a classification model to distinguish summer crops from other landforms; S4: Model training and verification: Model training and verification are performed on the GEE platform to ensure that the model has good generalization capabilities.

10. The remote sensing identification method and system for abandoned sloping farmland according to claim 5, characterized in that: The remote sensing identification method sets different abandonment scenarios under different scenarios to estimate the potential future food loss and its impact on food security. The specific steps are as follows: S1: Scenario setting: Different abandonment scenarios are set based on historical data analysis and policy assumptions, such as complete abandonment, partial abandonment and no abandonment; S2: Calculation of food loss: The food loss under each scenario is calculated using the formula Qp = Sap × a × Yieldp, where Qp is the loss, Sap is the affected cultivated land area, a is the yield reduction ratio, and Yieldp is the original yield; S3: Risk assessment: Comprehensively assess food security risks under various scenarios by combining regional population growth and economic development factors; S4: Policy recommendation formulation: Propose corresponding agricultural management and land protection policy recommendations based on the results of scenario analysis.

Images