Landslide hidden danger grading and risk assessment method and system considering surface deformation

Abstract

The invention discloses a landslide hidden danger grading and risk assessment method and system considering earth surface deformation, and the method comprises the steps: S1, extracting index data corresponding to disaster-inducing factors, disaster-pregnancy factors and disaster-bearing factors, generating a landslide hidden danger risk assessment factor database, and constructing a landslide hidden danger risk assessment factor system; s2, generating a potential landslide slippage range; s3, determining hierarchical relationships of the landslide hidden danger risk evaluation factor system, including a target layer-factor layer and a factor layer-index layer, and solving weight coefficients of the two hierarchical relationships; and S4, carrying out standardization and weighted summation on the numerical value of each index to obtain a landslide hidden danger risk score value, and classifying the landslide hidden danger risk score value to obtain a landslide hidden danger risk grade. According to the grading method for the potential landslide hidden dangers, the disaster-inducing factors, the disaster-pregnancy factors and the disaster-bearing factors are integrated, and technical support is provided for rapid screening of wide-area high and medium-prone landslide hidden dangers.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence, remote sensing and geological disaster identification and remote sensing monitoring technology, and in particular to a landslide hazard classification and risk assessment method and system taking into account surface deformation. Background Art

[0002] Landslides are geological disasters that have a serious impact on my country's economic and social development. my country has complex geological conditions and frequent tectonic activities. Landslides are a common disaster with many hidden dangers, wide distribution, and strong suddenness, causing heavy casualties and economic and property losses every year. Once a landslide occurs, its impact range is large. The collapse, mudslides and other sliding movements it causes will seriously threaten the people, houses, roads and other infrastructure located on the sliding path of the landslide body. Therefore, it is of great significance to quickly conduct landslide risk assessment in advance based on the distribution of the disaster-bearing body before the disaster occurs, which is of great significance for landslide disaster monitoring and prevention.

[0003] There are three issues to be addressed in the classification and risk assessment of landslide hazards. The first is where the surface deformation and the surrounding hazard-bearing bodies are, the second is how large the potential landslide sliding range is, and the third is how to classify suspected landslide hazards according to the degree of danger. In the past, the identification of hazard-bearing bodies was mainly based on manual interpretation of high-resolution optical data, but with the development of computer and artificial intelligence technology, the identification of hazard-bearing bodies based on deep neural networks has gradually become a mainstream trend. The surface deformation is mainly extracted through InSAR data, and the abnormal deformation areas are identified based on manual interpretation, threshold segmentation or cluster analysis. Since there is a lot of noise information in the InSAR surface deformation phase data and the causes of deformation are multi-solution, the above methods are difficult to achieve efficient and accurate identification. The current landslide sliding range simulation methods are mainly divided into empirical methods and modeling methods. However, the empirical method has low prediction accuracy and relies on existing statistical data. There is a lack of historical statistical data on large-scale landslides, making it difficult to effectively apply it to engineering scenarios. The modeling method needs to consider the complex local geological conditions, and the calculation process relies on complete geotechnical mechanics parameters. However, in actual engineering tasks, there is a lack of reliable relevant parameters, and it is impossible to quickly generate the landslide sliding range on a large scale. Therefore, it is difficult to achieve good results when the above methods are directly applied to the potential landslide disaster risk assessment business work. Summary of the invention

[0004] In view of the deficiencies in the prior art, the present invention provides a landslide hazard classification and risk assessment method taking into account surface deformation, the method comprising:

[0005] S1. Based on radar images and terrain data, by integrating InSAR technology, deep learning technology, and GIS spatial analysis technology, extract the index data corresponding to the disaster-causing factors, disaster-forming factors, and disaster-bearing factors, as well as the surface deformation phase and the InSAR surface deformation phase anomaly area, generate a landslide hazard risk assessment factor database, and construct a landslide hazard risk assessment factor system; among them, the disaster-causing factors include the surface deformation rate, and the corresponding disaster-causing factor index includes the average surface deformation rate. The disaster-forming factors include slope, elevation, and slope unit, and the corresponding disaster-forming factor indexes include the maximum hidden danger height difference, average slope, and the number of slope units where it is located. The disaster-bearing factors include buildings, roads, and rivers, and the corresponding disaster-bearing factor indexes include the number of threatened buildings, the number of threatened roads, and the area of threatened water bodies;

[0006] S2. Based on the InSAR surface deformation phase anomaly area, the surface deformation phase, the surface deformation rate, the slope, and the elevation, determine the potential initial slip point grid, construct a Monte Carlo landslide random walk algorithm for slopes, design a slip truncation condition, and conduct Monte Carlo slope slip simulation until the slip truncation condition is reached. Count the frequency of each grid being selected in the slip path, convert the frequency data into a spatial vector, and generate the potential landslide slip range, which is used as the potential threat range of landslide hazards to extract the index data of hazard-related disaster-bearing factors through GIS spatial analysis technology;

[0007] S3. Use the analytic hierarchy process to determine the hierarchical relationship of the landslide hazard risk assessment factor system, including the target layer - factor layer and the factor layer - index layer, construct a pairwise comparison matrix, and combine expert opinions and relevant materials to solve the weight coefficients of the two hierarchical relationships. Among them, the target layer is the landslide hazard risk score value, the factor layer includes disaster-causing factors, disaster-forming factors, and disaster-bearing factors, and the index layer includes disaster-causing factor indexes, disaster-forming factor indexes, and disaster-bearing factor indexes;

[0008] S4. Standardize the values of each index, and based on the weight coefficients of the two hierarchical relationships, perform weighted summation on the standardized values to obtain the landslide hazard risk score value, and classify the landslide hazard risk score value to obtain the landslide hazard risk level.

[0009] Preferably, in step S1, the typical disaster-bearing body targets, that is, the disaster-bearing factors, are extracted by the following method, and the deformation phase data and elevation data are used as the input for model training and inference. The InSAR surface deformation phase anomaly area is extracted by the following method:

[0010] (1) Collect sample data

[0011] Collect high-resolution remote sensing images of the study area as basic data, select targets with representative features for manual interpretation as reference truth values, combine the marked truth value images and the corresponding original high-resolution images in bands, cut the combined images into slices of a preset size, and randomly divide them into a sample training set and a sample test set according to a preset ratio;

[0012] (2) Build a model

[0013] Build a multi-scale feature fusion network model. This model adopts an encoder-decoder architecture. The encoder part gradually extracts the features of the image through multiple Depconvgroup modules. Each Depconvgroup module consists of two groups of depthwise separable convolutional layers, a BN normalization layer, and a Relu activation layer. The decoder part consists of Cbri modules of different scales. Each Cbri module consists of a convolutional layer, a BN normalization layer, and an upsampling layer. After upsampling, the features of different scales are processed into a unified size. Finally, the features of different levels are concatenated in the channel dimension, and a channel attention mechanism module is introduced to enhance the model's ability to extract key features;

[0014] (3) Train the model

[0015] Use the slice data of the sample training set to train the multi-scale feature fusion network model for multiple rounds. If the error index decreases and tends to be stable, it means that the model training is completed. Use the slice data in the sample test set to verify the multi-scale feature fusion network model. If the verification model accuracy does not meet the requirements, adjust the parameters of the model training and retrain until the accuracy requirements are met;

[0016] (4) Model inference

[0017] Use the trained multi-scale feature fusion network model to predict all the data in the study area to obtain the distribution results of typical disaster-bearing bodies in the whole area.

[0018] Preferably, in step S2, build a ramp Monte Carlo landslide random walk algorithm, including:

[0019] Take the InSAR surface deformation phase anomaly area as the initial potential slip point of the landslide, and build a random probability calculation formula for the current slip grid to slip to the target grid. The calculation formula is:

[0020]

[0021] Among them, P px is the probability that the target grid around the current slip grid becomes the next slip grid, n is the number of target grids, β is the slope between the current slip grid and the target grid, f d and fβ They are weight factors for adjusting the sliding direction and slope. When the direction of the next sliding grid is the same as the previous one, f d is d 2 , when the sliding direction changes by 45°, f d is d, when the sliding direction changes by 90°, f d is 1, f β and d are both initial input parameters, f r is the average surface deformation phase rate;

[0022] Estimate the maximum horizontal sliding distance as the sliding cut-off condition, and the calculation formula is:

[0023] L = K1×ΔH + K2(f s + f r ) + B' (3)

[0025] where L is the maximum horizontal sliding distance, ΔH is the height difference from the top of the landslide to the ground, f s and f r are the area of the surface deformation phase aggregation area and the average surface deformation phase rate respectively, K1 and K2 are correlation coefficients, B' is a constant term, and K1, K2 and B' can be obtained by fitting historical observation data, and the values are 2.54, 0.5 and -50.37 respectively;

[0026] Statistically calculate the frequency of each grid in the sliding path being selected by the sliding simulation, that is, calculate the probability of sliding to the target grid each time. The calculation formula is:

[0027]

[0028] where P is the influence probability of the landslide sliding on the target grid, s i is the number of times the sliding point selects the target grid during the simulation process, m is the number of hidden dangers to which the sliding point affecting the target grid belongs, n max is the maximum number of iterations of the Monte Carlo random walk simulation, and i and m are positive integers.

[0029] Preferably, in step S3, the analytic hierarchy process is used to solve the hierarchical relationship weights of the target layer - factor layer and the factor layer - index layer respectively;

[0030] Using the analytic hierarchy process to solve the weight of a single hierarchical relationship includes:

[0031] Construct a matrix M, and the elements in the matrix are marked as a ij , combined with expert opinions and relevant materials, compare the evaluation indicators of the two levels to be solved pairwise, and fill the determined scale into the corresponding position of the matrix M to obtain the matrix M, which is the pairwise comparison matrix; where a ijIndicates the importance or scale of i compared to j, a ij The corresponding position is the element in the i-th row and j-th column of matrix M; if i = j, it means that i and j are equally important, that is, a ij = 1, and the scales of the main diagonal elements of matrix M are all 1; a ij > 0 and a ij ×a ji = 1;

[0032] The arithmetic mean method, geometric mean method, and eigenvalue method are respectively used to calculate the weights, and the arithmetic mean of the calculation results is obtained. Assume that the pairwise comparison matrix M is:

[0033]

[0034] The weight value w of the i-th factor is obtained through the following three methods i , where

[0035] First, the weight vector w i obtained by the arithmetic mean method is:

[0036]

[0037] Second, the weight vector w i obtained by the geometric mean method is:

[0038]

[0039] Third, for the eigenvalue method to calculate the weight w i , first assume that the largest eigenvalue of the pairwise comparison matrix M is λ max , and its corresponding eigenvector is ξ. Then the weight w i obtained by the eigenvalue method is:

[0040]

[0041] The pairwise comparison matrix is subjected to consistency test through the following methods, including:

[0042] Calculate the largest eigenvalue λ of the eigenvector of the pairwise comparison matrix max , and the calculation formula is:

[0043]

[0044] where M is the pairwise comparison matrix, w i is the index weight, W = [w1,…,w n T , n is the order of the comparison matrix, and i, j, k, n are positive integers;

[0045] Calculate the consistency index CI of the pairwise comparison matrix, and the calculation formula is:​

[0046]

[0047] When the CI is close to 0, it is initially judged that the consistency test of the constructed pairwise comparison matrix passes;

[0048] Calculate the consistency ratio, and the calculation formula is:

[0049]

[0050] Among them, RI is the average consistency index, which is the parameter value obtained from the lookup table. Usually, when CR < 0.1, it means that the consistency test of the pairwise comparison matrix passes.

[0051] Preferably, in step S4, the Z-Score method is used to standardize and weighted sum the values of each index, and the calculation formula is:

[0052]

[0053] Among them, R is the risk score value after comprehensively considering the risk evaluation index, w j is the j-th factor, m is the number of factor layers, w i is the weight of the i-th index, n is the number of indexes in the factor layer, x i is the actual observed value or measured value of the i-th index, μ is the average value of all observed values, σ is the standard deviation of all index data, representing the degree of data dispersion, and i, j, m, n are positive integers.

[0054] Preferably, in step S4, the k-means algorithm is used to classify the landslide hidden danger risk score value, including:

[0055] Pre-determine the number of classification categories K, and randomly select K objects {a1, a2,..., a K} as the initial clustering centers of the landslide hidden danger risk score value data {X1, X2,..., X n}, calculate the distance between each data and each clustering center, assign it to each clustering center according to the principle of the nearest distance, and recalculate the center of each category, finally making the sum of the squares of the distances from each data point to the nearest clustering center the smallest. The calculation formula is:

[0056]

[0057] Among them, T is the distance objective function of clustering, q j is the j-th initial clustering center, K is the number of classification categories, X i is the data to be clustered, n is the number of data, and i, j, n are positive integers.

[0058] Based on the same inventive concept, the present invention further provides a landslide hazard grading and risk assessment system considering surface deformation, and the system includes:

[0059] A construction module, configured to extract index data corresponding to disaster-causing factors, disaster-forming factors, and disaster-bearing factors, as well as surface deformation phase and InSAR surface deformation phase anomaly areas, generate a landslide hazard risk evaluation factor database, and construct a landslide hazard risk evaluation factor system based on radar images and terrain data by integrating InSAR technology, deep learning technology, and GIS spatial analysis technology; wherein, the disaster-causing factors include surface deformation rate, and the corresponding disaster-causing factor index includes average surface deformation rate, the disaster-forming factors include slope, elevation, and slope unit, and the corresponding disaster-forming factor indexes include maximum hidden danger height difference, average slope, and number of slope units where it is located, and the disaster-bearing factors include buildings, roads, and rivers, and the corresponding disaster-bearing factor indexes include number of threatened buildings, number of threatened roads, and threatened water body area;

[0060] A generation module, configured to determine potential initial slip point grids based on the InSAR surface deformation phase anomaly area, the surface deformation phase, the surface deformation rate, the slope, and the elevation, construct a Monte Carlo landslide random walk algorithm for slopes, design a slip truncation condition, perform Monte Carlo slope slip simulation until the slip truncation condition is reached, count the frequencies of each grid selected in the slip path, convert the frequency data into a spatial vector, generate a potential landslide slip range, which is used as the potential threat range of landslide hazards, and extract index data of hazard-related disaster-bearing factors through GIS spatial analysis technology;

[0061] A solution module, configured to determine the hierarchical relationship of the landslide hazard risk evaluation factor system by using the analytic hierarchy process, including the target layer - factor layer and the factor layer - index layer, construct a pairwise comparison matrix, and solve the weight coefficients of the two hierarchical relationships in combination with expert opinions and relevant materials, wherein the target layer is the landslide hazard risk score value, the factor layer includes disaster-causing factors, disaster-forming factors, and disaster-bearing factors, and the index layer includes disaster-causing factor indexes, disaster-forming factor indexes, and disaster-bearing factor indexes;

[0062] An evaluation module, configured to standardize the numerical values of each index, perform weighted summation on the standardized numerical values according to the weight coefficients of the two hierarchical relationships to obtain the landslide hazard risk score value, and classify the landslide hazard risk score value to obtain the landslide hazard risk level.

[0063] Preferably, the following method is used to extract typical disaster-bearing body targets, i.e., disaster-bearing factors, and the deformation phase data and elevation data are used as the input for model training and inference. The following method is used to extract the InSAR surface deformation phase anomaly area:

[0064] (1) Collect sample data

[0065] Collect high-resolution remote sensing images of the study area as basic data, select representative targets for manual interpretation as reference true values, combine the marked true value images with the corresponding original high-resolution images in bands, cut the combined images into slices of a preset size, and randomly divide them into a sample training set and a sample test set according to a preset ratio;

[0066] (2) Build a model

[0067] Build a multi-scale feature fusion network model. This model adopts an encoder-decoder architecture. In the encoder part, the features of the image are gradually extracted through multiple Depconvgroup modules. Each Depconvgroup module consists of two groups of depthwise separable convolutional layers, a BN normalization layer, and a Relu activation layer. The decoder part consists of Cbri modules of different scales. Each Cbri module consists of a convolutional layer, a BN normalization layer, and an upsampling layer. After upsampling, the features of different scales are processed into a unified size. Finally, the features of different levels are concatenated in the channel dimension, and a channel attention mechanism module is introduced to enhance the model's ability to extract key features;

[0068] (3) Train the model

[0069] Use the slice data of the sample training set to train the multi-scale feature fusion network model for multiple rounds. If the error index decreases and tends to be stable, it means that the model training is completed. Use the slice data in the sample test set to verify the multi-scale feature fusion network model. If the verification model accuracy does not meet the requirements, adjust the parameters of the model training and retrain until the accuracy requirements are met;

[0070] (4) Model inference

[0071] Use the trained multi-scale feature fusion network model to predict all the data in the study area to obtain the distribution results of typical disaster-bearing bodies in the whole area.

[0072] Preferably, build a ramp Monte Carlo landslide random walk algorithm, including:

[0073] Take the InSAR surface deformation phase anomaly area as the initial potential slip point of the landslide, and build a random probability calculation formula for the current slip grid to slip to the target grid. The calculation formula is:

[0074]

[0075] where P px is the probability that the target grid around the current slip grid becomes the next slip grid, n is the number of target grids, β is the slope between the current slip grid and the target grid, fd and f β are the weight factors for adjusting the sliding direction and slope respectively. When the direction of the next sliding grid is the same as the previous one, f d is d 2 , when the sliding direction changes by 45°, f d is d, when the sliding direction changes by 90°, f d is 1, f β and d are both initialization input parameters, f r is the average surface deformation phase rate;

[0076] Estimate the maximum horizontal sliding distance as the sliding cut-off condition, and the calculation formula is:

[0077] L = K1×ΔH + K2(f s + f r ) + B' (3)

[0079] where L is the maximum horizontal sliding distance, ΔH is the height difference from the top of the landslide to the ground, f s and f r are the area of the surface deformation phase aggregation area and the average surface deformation phase rate respectively, K1 and K2 are correlation coefficients, B' is a constant term, and K1, K2 and B' can be obtained by fitting historical observation data, and the values are 2.54, 0.5 and -50.37 respectively;

[0080] Statistically calculate the frequency of each grid in the sliding path being selected by the sliding simulation, that is, calculate the probability of sliding to the target grid each time. The calculation formula is:

[0081]

[0082] where P is the influence probability of the landslide sliding on the target grid, s i is the number of times the sliding point selects the target grid in the simulation process, m is the number of hidden dangers to which the sliding point affecting the target grid belongs, n max is the maximum number of iterations of the Monte Carlo random walk simulation, and i and m are positive integers.

[0083] Preferably, the analytic hierarchy process is used to solve the hierarchical relationship weights of the target layer - factor layer and the factor layer - index layer respectively;

[0084] Using the analytic hierarchy process to solve the weight of a single hierarchical relationship includes:

[0085] Construct a matrix M, and the elements in the matrix are marked as a ij , combined with expert opinions and relevant materials, compare the evaluation indicators of the two levels to be solved pairwise, and fill the determined scale into the corresponding position of the matrix M to obtain the matrix M, which is the pairwise comparison matrix; where aij Indicates the importance or scale of i compared to j, a ij The corresponding position is the element in the i-th row and j-th column of matrix M; if i = j, it means i and j are equally important, that is, a ij = 1, and the scale of the main diagonal elements of matrix M is all 1; a ij > 0 and a ij × a ji = 1;

[0086] The arithmetic mean method, geometric mean method, and eigenvalue method are respectively used to calculate the weights, and the arithmetic mean of the calculation results is obtained. Assume the pairwise comparison matrix M is:

[0087]

[0088] The weight value w of the i-th factor is obtained through the following three methods i , where

[0089] First, the arithmetic mean method is used to obtain the weight vector w i as:

[0090]

[0091] Second, the geometric mean method is used to obtain the weight vector w i as:

[0092]

[0093] Third, for the eigenvalue method to obtain the weight w i , first assume that the largest eigenvalue of the pairwise comparison matrix M is λ max , and its corresponding eigenvector is ξ. Then the weight w i obtained by the eigenvalue method is:

[0094]

[0095] The pairwise comparison matrix is subjected to consistency test through the following methods, including:

[0096] Calculate the largest eigenvalue λ of the eigenvector of the pairwise comparison matrix max , and the calculation formula is:

[0097]

[0098] where M is the pairwise comparison matrix, w i is the index weight, W = [w1,..., w n T , n is the order of the comparison matrix, and i, j, k, n are positive integers;

[0099] ​Calculate the consistency index CI of the pairwise comparison matrix, the calculation formula is:

[0100]

[0101] When CI is close to 0, it is preliminarily judged that the consistency test of the constructed pairwise comparison matrix has passed;

[0102] Calculate the consistency ratio using the following formula:

[0103]

[0104] Among them, RI is the average consistency index, the parameter value obtained from the lookup table. Usually, when CR<0.1, it means that the consistency test of the pairwise comparison matrix has passed.

[0105] Compared with the prior art, the present invention has the following beneficial effects:

[0106] The present invention uses deep learning technology and multi-source remote sensing data to accurately identify the InSAR surface deformation anomaly area and accurately locate potential landslide hazards; draws on the idea of ​​random walk to propose an automatic prediction method for the sliding range of potential landslides that takes into account deformation characteristics, providing technical support for the risk assessment of landslide hazards on a wide scale; proposes a classification method for potential landslide hazards that integrates disaster-causing factors, disaster-pregnant factors, and disaster-bearing factors, providing technical support for the rapid screening of high-risk landslide hazards in a wide area. The technology of the present invention plays an important role in improving the automatic identification of potential landslide hazards in a wide area, and is of great significance for improving the intelligent level of landslide disaster monitoring and early warning. BRIEF DESCRIPTION OF THE DRAWINGS

[0107] Figure 1 A schematic flow chart of a landslide hazard classification and risk assessment method taking into account surface deformation provided by the present invention;

[0108] Figure 2 A technical flow chart of a landslide hazard classification and risk assessment method taking into account surface deformation provided by the present invention;

[0109] Figure 3 Various hidden danger risk assessment factors provided by the present invention;

[0110] Figure 4 A structural framework diagram of the multi-scale feature fusion network model provided by the present invention;

[0111] Figures 5-6 A schematic diagram of the InSAR surface deformation phase anomaly area identification result provided by the present invention;

[0112] Figure 7 This is a schematic diagram of the potential landslide risk assessment results in Zhouqu County provided by the present invention. Specific Embodiments

[0113] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0114] The present invention will be further described in detail below with reference to the accompanying drawings.

[0115] As Figures 1-2 shown, the embodiments of the present invention provide a landslide hazard grading and risk assessment method considering surface deformation, including:

[0116] S1. Based on radar images and terrain data, by integrating InSAR technology, deep learning technology, and GIS spatial analysis technology, extract the index data corresponding to the disaster-causing factors, disaster-forming factors, and disaster-bearing factors, as well as the surface deformation phase and the InSAR surface deformation phase anomaly area, generate a landslide hazard risk assessment factor database, and construct a landslide hazard risk assessment factor system; wherein, the disaster-causing factors include the surface deformation rate, and the corresponding disaster-causing factor index includes the average surface deformation rate, the disaster-forming factors include slope, elevation, and slope unit, and the corresponding disaster-forming factor indexes include the maximum height difference of the hidden danger, the average slope, and the number of slope units where it is located, and the disaster-bearing factors include buildings, roads, and rivers, and the corresponding disaster-bearing factor indexes include the number of threatened buildings, the number of threatened roads, and the area of threatened water bodies;

[0117] Step 1: Construct a landslide hazard risk assessment factor system

[0118] The purpose of this step is to establish a comprehensive landslide hazard risk assessment factor system covering three aspects: disaster-causing factors, disaster-forming factors, and disaster-bearing bodies. Specifically, select the maximum height difference of the hidden danger, the average slope, and the number of slope units where it is located as the disaster-forming factor indexes, the number of threatened buildings, roads, and the water area as the disaster-bearing factor indexes, and the average surface deformation rate detected based on InSAR technology as the disaster-causing factor index. By integrating InSAR technology, deep learning technology, and GIS spatial analysis technology, extract the corresponding index data, and generate a landslide hazard risk assessment factor database ( Figure 3 ), providing basic data for subsequent risk assessment. Figure 3 Shows various hazard risk assessment factors, (a) altitude, (b) slope, (c) deformation rate, (d) population density, (e) slope unit, (f) building, (g) road, (h) water body.

[0119] In the embodiments of the present invention, the typical disaster-bearing body targets, i.e., disaster-bearing factors, are extracted by the following method. Moreover, deformation phase data and elevation data are used as the input for model training and inference, and the InSAR surface deformation phase anomaly area is extracted by the following method:

[0120] (1) Collect sample data

[0121] Collect high-resolution remote sensing images of the study area as basic data, select targets with representative features for manual interpretation as reference truth values, combine the marked truth value images and the corresponding original high-resolution images in bands, cut the combined images into several slices of a preset size (256×256), and randomly divide them into a sample training set and a sample test set according to a preset ratio (7:3);

[0122] (2) Build a model

[0123] Build a multi-scale feature fusion network model (MFAFNet, Multi-Level Feature Attention Fusion Network) for information extraction. As Figure 4 shown, this model adopts an encoder-decoder architecture. The encoder part gradually extracts the features of the image through multiple Depconvgroup modules. Each Depconvgroup module consists of two groups of depthwise separable convolutional layers, a BN normalization layer, and a Relu activation layer, with less computational complexity compared to traditional two-dimensional convolutional layers. The decoder part consists of Cbri modules of different scales. Each Cbri module consists of a convolutional layer, a BN normalization layer, and an upsampling layer (UpSample). After upsampling, the features of different scales are processed into a unified size. Finally, the features of different levels are concatenated (Cat) in the channel dimension, and a channel attention mechanism module (SE module) is introduced to improve the model's ability to extract key features. Thus, the model construction is completed; Figure 4 In, Image represents the image input to the model, H represents the image height, W represents the image width, MLP (multi-layer perceptron) represents a multi-layer perceptron, which is a feedforward neural network model, and Prediction represents the prediction result.

[0124] (3) Train the model

[0125] Use the slice data of the sample training set to train the multi-scale feature fusion network model for multiple rounds. If the error index decreases and tends to be stable, it means that the model training is completed. Use the slice data in the sample test set to verify the multi-scale feature fusion network model. If the verification model accuracy does not meet the requirements, adjust the model training parameters and retrain until the accuracy requirements are met;

[0126] (4) Model inference

[0127] Use the trained multi-scale feature fusion network model to predict all the data in the study area, and obtain the distribution results of typical disaster-bearing bodies in the whole area. Figure 5 The identification results of the surface deformation phase anomaly area are shown.

[0128] S2. Based on InSAR surface deformation phase anomaly area, surface deformation phase, surface deformation rate, slope, and elevation, determine the grid of potential initial slip points, construct a Monte Carlo landslide random walk algorithm for slopes, design a slip truncation condition, conduct Monte Carlo slope slip simulation until the slip truncation condition is reached, count the frequency of each grid being selected in the slip path, convert the frequency data into a spatial vector, generate the potential landslide slip range, which is used as the potential threat range of landslide hazards, and extract the index data of disaster-bearing factors related to hazards through GIS spatial analysis technology;

[0129] Step 2: Estimation of landslide influence range

[0130] Determine the potential landslide starting slip area, construct a Monte Carlo landslide random walk algorithm for slopes, and simulate the longitudinal and lateral movement processes of the slip. Design a slip truncation condition, conduct a large number of random walks for the slip starting point until the truncation distance condition is reached. Finally, count the frequency of each grid being selected in the slip path, convert the probability data into a spatial vector, and realize the automatic generation of the potential landslide slip range. The slip result generated in this step is used as the potential threat range of landslide hazards, and is used to extract the index data of disaster-bearing factors related to hazards through geospatial analysis technology.

[0131] In the embodiment of the present invention, constructing a Monte Carlo landslide random walk algorithm for slopes includes:

[0132] Take the InSAR surface deformation phase anomaly area as the initial potential slip point of the landslide, and construct a random probability calculation formula for the current slip grid to slip to the target grid. The calculation formula is:

[0133]

[0134]

[0135] Among them, P px is the probability that the target grid around the current slip grid becomes the next slip grid, n is the number of target grids, β is the slope between the current slip grid and the target grid, f d and f β are the weight factors for adjusting the sliding direction and slope respectively. When the direction of the next slip grid is the same as the previous one, f d is d 2 ; when the sliding direction changes by 45°, fd is d. When the sliding direction changes by 90°, f d is 1, f β and d are both initialization input parameters, f r is the average surface deformation phase rate;

[0136] Estimate the maximum horizontal sliding distance as the sliding cut-off condition, and the calculation formula is:

[0137] L = K1×ΔH + K2(f s + f r ) + B' (3)

[0139] where L is the maximum horizontal sliding distance, ΔH is the height difference from the top of the landslide to the ground, f s and f r are the area of the surface deformation phase aggregation area and the average surface deformation phase rate respectively, K1 and K2 are correlation coefficients, B' is a constant term, and K1, K2 and B' can be obtained by fitting historical observation data, and the values are 2.54, 0.5 and -50.37 respectively;

[0140] Statistically analyze the frequency of each grid in the sliding path being selected by the sliding simulation, that is, calculate the probability of sliding to the target grid each time. The calculation formula is:

[0141]

[0142] where P is the influence probability of the landslide sliding on the target grid, s i is the number of times the sliding point selects the target grid during the simulation process, m is the number of hidden dangers to which the sliding point affecting the target grid belongs, n max is the maximum number of iterations of the Monte Carlo random walk simulation, and i, m are positive integers.

[0143] S3. Use the analytic hierarchy process to determine the hierarchical relationship of the landslide hidden danger risk assessment factor system, including the target layer - factor layer and the factor layer - index layer, construct a pairwise comparison matrix, and combine expert opinions and relevant materials to solve the weight coefficients of the two hierarchical relationships. Among them, the target layer is the landslide hidden danger risk score value, the factor layer includes disaster-causing factors, disaster-forming factors and disaster-bearing factors, and the index layer includes disaster-causing factor indicators, disaster-forming factor indicators and disaster-bearing factor indicators;

[0144] Step three: Determine the hierarchical relationship of the evaluation system and calculate the weights

[0145] After constructing the evaluation factor system, the hierarchical relationship of the evaluation system is determined by the analytic hierarchy process, including the target layer, factor layer, and index layer. The target layer is the landslide hazard risk score, the factor layer includes disaster-causing factors, disaster-bearing factors, and disaster-triggering factors, and the index layer is the specific evaluation indicators selected in step one. By constructing a pairwise comparison matrix and combining expert opinions and relevant data, pairwise comparisons are made for each factor, and the weight coefficients of the factor layer and index layer are solved.

[0146] In the embodiment of the present invention, the analytic hierarchy process is used to solve the hierarchical relationship weights of the target layer - factor layer and factor layer - index layer respectively;

[0147] Using the analytic hierarchy process to solve the weight of a single hierarchical relationship includes:

[0148] Construct a matrix M, and the elements in the matrix are marked as a ij , combining expert opinions and relevant data, make pairwise comparisons of the evaluation indicators of the two levels to be solved, and fill the determined scale into the corresponding position of the matrix M to obtain the pairwise comparison matrix M; among them, a ij represents the importance or scale of i compared with j, and the position corresponding to a ij is the i-th row and j-th column in the matrix M; if i = j, it means that i and j are equally important, that is, a ij = 1, and the scales of the main diagonal elements of the matrix M are all 1; a ij > 0 and a ij ×a ji = 1;

[0149] Calculate the weights respectively by the arithmetic mean method, geometric mean method, and eigenvalue method, and calculate the arithmetic mean of the calculation results. Assume that the pairwise comparison matrix M is:

[0150]

[0151] The weight value w i of the i-th factor is obtained through the following three methods, where

[0152] First, the weight vector w i obtained by the arithmetic mean method is:

[0153]

[0154] Second, the weight vector w i obtained by the geometric mean method is:

[0155]

[0156] Third, for the eigenvalue method to find the weight w i , first assume that the largest eigenvalue of the pairwise comparison matrix M is λ max, its corresponding eigenvector is ξ, and the weight w is obtained by the eigenvalue method i is:

[0157]

[0158] Since the pairwise comparison matrix is determined based on expert scoring, a consistency test is required to ensure that there are no conflicting inputs in the pairwise comparison matrix. The pairwise comparison matrix is subjected to a consistency test through the following methods, including:

[0159] Calculate the maximum eigenvalue λ of the eigenvector of the pairwise comparison matrix max , and the calculation formula is:

[0160]

[0161] where M is the pairwise comparison matrix, w i is the index weight, W = [w1,…,w n T , n is the order of the comparison matrix, and i, j, k, n are positive integers;

[0162] Calculate the consistency index CI of the pairwise comparison matrix, and the calculation formula is:

[0163]

[0164] When CI is close to 0, it is initially judged that the consistency test of the constructed pairwise comparison matrix passes;

[0165] Calculate the consistency ratio, and the calculation formula is:

[0166]

[0167] where RI is the average consistency index, a parameter value obtained from the lookup table. Usually, when CR < 0.1, it means that the consistency test of the pairwise comparison matrix passes.

[0168] S4. Standardize the values of each index. According to the weight coefficients of the two-level relationship, perform weighted summation on the standardized values to obtain the landslide hazard risk score value, and classify the landslide hazard risk score value to obtain the landslide hazard risk level.

[0169] Step 4: Landslide hazard risk level division

[0170] Use the Z-score standardization method to unify the value range of each index to 0-1. According to the obtained weight coefficients, perform weighted summation on the standardized values to calculate the landslide hazard risk score. Use the k-means clustering algorithm to classify the calculated risk score values to complete the division of the risk level. ​

[0171] In the embodiments of the present invention, the Z-Score method is used to standardize and sum the weighted values of each index, and the calculation formula is as follows:

[0172]

[0173]

[0174] Among them, R is the risk score value after comprehensively considering risk evaluation indicators, w j is the jth factor, m is the number of factor layers, w i is the weight of the ith index, n is the number of indicators in the factor layer, x i is the actual observed value or measured value of the ith index, μ is the average value of all observed values, σ is the standard deviation of all index data, indicating the degree of data dispersion, and i, j, m, n are positive integers.

[0175] In the embodiments of the present invention, the k-means algorithm is used to classify the landslide hidden danger risk score values, including:

[0176] Pre-determine the number of classification categories K, and randomly select K objects {a1, a2,..., a K} as the initial clustering centers of the landslide hidden danger risk score value data {X1, X2,..., X n}, calculate the distance between each data and each clustering center, assign it to each clustering center according to the principle of the nearest distance, and recalculate the center of each category, finally making the sum of the squares of the distances from each data point to the nearest clustering center the smallest. The calculation formula is as follows:

[0177]

[0178] Among them, T is the distance objective function of clustering, q j is the jth initial clustering center, K is the number of classification categories, X i is the data to be clustered, n is the number of data, and i, j, n are positive integers.

[0179] Taking the Zhouqu area in Gansu as an example, the evaluation work of landslide hidden dangers is carried out. The risk levels are set to 5 categories, namely low, relatively low, medium, relatively high, and high. According to the k-means clustering algorithm, the threshold of each category is calculated (Table 1), and the recognition results are as Figure 7 shown.

[0180] Table 1 Threshold table of landslide hidden danger risk evaluation categories

[0181] Risk category Numerical range Low 0～0.187 Lower 0.187～0.236 Medium 0.236～0.299 Higher 0.299～0.395 High 0.395～0.496

[0182] Based on the same inventive concept, an embodiment of the present invention further provides a landslide hazard grading and risk assessment system considering surface deformation, including:

[0183] A construction module, configured to extract index data corresponding to disaster-causing factors, disaster-forming factors, and disaster-bearing factors, as well as surface deformation phase and InSAR surface deformation phase anomaly areas, based on radar images and terrain data, by integrating InSAR technology, deep learning technology, and GIS spatial analysis technology, generate a landslide hazard risk assessment factor database, and construct a landslide hazard risk assessment factor system; wherein, the disaster-causing factors include surface deformation rate, and the corresponding disaster-causing factor index includes average surface deformation rate, the disaster-forming factors include slope, elevation, and slope unit, and the corresponding disaster-forming factor indexes include maximum hidden danger height difference, average slope, and number of slope units where it is located, and the disaster-bearing factors include buildings, roads, and rivers, and the corresponding disaster-bearing factor indexes include number of threatened buildings, number of threatened roads, and threatened water area;

[0184] A generation module, configured to determine potential initial slip point grids based on the InSAR surface deformation phase anomaly area, surface deformation phase, surface deformation rate, slope, and elevation, construct a Monte Carlo landslide random walk algorithm for slopes, design a slip truncation condition, perform Monte Carlo slope slip simulation until the slip truncation condition is reached, count the frequencies of each grid selected in the slip path, convert the frequency data into a spatial vector, generate a potential landslide slip range, which is used as the potential threat range of landslide hazards, for extracting index data of associated disaster-bearing factors of hazards through GIS spatial analysis technology;

[0185] A solution module, configured to determine the hierarchical relationship of the landslide hazard risk assessment factor system by using the analytic hierarchy process, including the target layer - factor layer and the factor layer - index layer, construct a pairwise comparison matrix, and solve the weight coefficients of the two hierarchical relationships in combination with expert opinions and relevant materials, wherein the target layer is the landslide hazard risk score, the factor layer includes disaster-causing factors, disaster-forming factors, and disaster-bearing factors, and the index layer includes disaster-causing factor indexes, disaster-forming factor indexes, and disaster-bearing factor indexes;

[0186] An evaluation module, configured to standardize the values of each index, perform weighted summation on the standardized values according to the weight coefficients of the two hierarchical relationships, obtain the landslide hazard risk score, and classify the landslide hazard risk score to obtain the landslide hazard risk level.

[0187] The present invention first uses a deep neural network and multi-source remote sensing data to complete the identification of typical disaster-bearing bodies and InSAR surface deformation phase anomaly areas, and determine the potential landslide slip starting area; according to data such as the InSAR surface deformation range, deformation rate, and elevation, generate a potential slip range through random walk simulation; and conduct comprehensive analysis in combination with the distribution of disaster-bearing bodies, surface deformation range and displacement, terrain, etc., to grade and risk-assess suspected landslide hazards.

[0188] Compared with the prior art, the advantages of the present invention are as follows:

[0189] 1. The present invention uses deep learning technology and multi-source remote sensing data to achieve accurate identification of InSAR surface deformation anomaly areas, and realizes the accurate positioning of potential landslide hazards.

[0190] 2. The present invention constructs a simulation method for the influence range of potential landslides based on the random walk algorithm, comprehensively considers constraint conditions such as sliding area, sliding rate, and elevation difference, solves the problem that traditional landslide influence range simulation methods are difficult to be extended to large-scale applications, and realizes the rapid and automatic prediction of the sliding influence range of potential landslides.

[0191] 3. The present invention comprehensively classifies potential landslide hazards by considering disaster-causing factors, disaster-bearing factors, and disaster-bearing factors, providing technical support for the rapid screening of potential landslide hazards with high to medium occurrence probability in a wide area.

[0192] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention can have various changes and modifications. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A landslide hazard grading and risk assessment method considering surface deformation, characterized in that The method includes the following steps: S1. Based on radar images and topographic data, by integrating InSAR technology, deep learning technology, and GIS spatial analysis technology, extract the index data corresponding to the disaster-causing factors, disaster-forming factors, and disaster-bearing factors, as well as the surface deformation phase and the InSAR surface deformation phase anomaly area, generate a landslide hazard risk assessment factor database, and construct a landslide hazard risk assessment factor system; wherein, the disaster-causing factors include the surface deformation rate, and the corresponding disaster-causing factor index includes the average surface deformation rate, the disaster-forming factors include slope, elevation, and slope unit, and the corresponding disaster-forming factor indexes include the maximum hidden danger height difference, average slope, and the number of slope units where it is located, and the disaster-bearing factors include buildings, roads, and rivers, and the corresponding disaster-bearing factor indexes include the number of threatened buildings, the number of threatened roads, and the area of threatened water bodies; S2. Based on the InSAR surface deformation phase anomaly area, the surface deformation phase, the surface deformation rate, the slope, and the elevation, determine the potential initial slip point grid, construct a Monte Carlo landslide random walk algorithm for slopes, design a slip truncation condition, conduct Monte Carlo slope slip simulation until the slip truncation condition is reached, count the frequencies of each grid selected in the slip path, convert the frequency data into a spatial vector, generate the potential landslide slip range, which is used as the potential threat range of landslide hazards, and extract the disaster-bearing factor index data through GIS spatial analysis technology; S3. Use the analytic hierarchy process to determine the hierarchical relationship of the landslide hazard risk assessment factor system, including the target layer - factor layer and the factor layer - index layer, construct a pairwise comparison matrix, and combine expert opinions and relevant data to solve the weight coefficients of the two hierarchical relationships, where the target layer is the landslide hazard risk score value, the factor layer includes disaster-causing factors, disaster-forming factors, and disaster-bearing factors, and the index layer includes disaster-causing factor indexes, disaster-forming factor indexes, and disaster-bearing factor indexes; S4. Standardize the values of each index, and according to the weight coefficients of the two hierarchical relationships, perform weighted summation on the standardized values to obtain the landslide hazard risk score value, and classify the landslide hazard risk score value to obtain the landslide hazard risk level.

2. The method according to claim 1, wherein In step S1, the typical disaster-bearing body targets, i.e., disaster-bearing factors, are extracted by the following method, and the deformation phase data and elevation data are used as the input for model training and inference. The InSAR surface deformation phase anomaly area is extracted by the following method: (1) Collect sample data Collect high-resolution remote sensing images of the study area as basic data, select targets with representative features for manual interpretation as reference truth values, combine the marked truth value images and the corresponding original high-resolution images in bands, cut the combined images into several slices of a preset size, and randomly divide them into a sample training set and a sample test set according to a preset ratio; (2) Construct a model Construct a multi-scale feature fusion network model. This model adopts an encoder-decoder architecture. In the encoder part, features of the image are gradually extracted through multiple Depconvgroup modules. Each Depconvgroup module consists of two sets of depthwise separable convolutional layers, a BN normalization layer, and a Relu activation layer. The decoder part is composed of Cbri modules of different scales. Each Cbri module consists of a convolutional layer, a BN normalization layer, and an upsampling layer. After upsampling, features of different scales are processed into a unified size. Finally, features of different levels are concatenated in the channel dimension, and a channel attention mechanism module is introduced to enhance the model's ability to extract key features; (3) Train the model Use the slice data of the sample training set to train the multi-scale feature fusion network model for multiple rounds. If the error index decreases and tends to be stable, it indicates that the model training is completed. Use the slice data of the sample test set to verify the multi-scale feature fusion network model. If the verification model accuracy does not meet the requirements, adjust the parameters of the model training and retrain until the accuracy requirement is met; (4) Model inference Use the trained multi-scale feature fusion network model to predict all the data in the study area to obtain the distribution results of typical disaster-bearing bodies in the whole area.

3. The method according to claim 1, wherein In step S2, construct a ramp Monte Carlo landslide random walk algorithm, including: Take the InSAR surface deformation phase anomaly area as the initial potential slip point of the landslide, and construct a random probability calculation formula for the current slip grid to slip to the target grid. The calculation formula is: Among them, P px is the probability that the target grid around the current slip grid becomes the next slip grid, n is the number of target grids, β is the slope between the current slip grid and the target grid, f d and f β are the weight factors for adjusting the sliding direction and slope respectively. When the direction of the next slip grid is the same as before, f d is d 2 . When the sliding direction changes by 45°, f d is d. When the sliding direction changes by 90°, f d is 1. f β and d are both initialization input parameters. f r is the average surface deformation phase rate; Estimate the maximum horizontal slip distance as the slip cut-off condition. The calculation formula is: L = K1×ΔH + K2f s + f r + B' (3) where L is the maximum horizontal sliding distance, ΔH is the height difference from the top of the landslide to the ground, f s and f r are the area of the surface deformation phase aggregation area and the average surface deformation phase rate respectively, K1 and K2 are correlation coefficients, B' is a constant term, and K1, K2 and B' can be obtained by fitting historical observation data, and the values are 2.54, 0.5 and -50.37 respectively; Statistically calculate the frequency of each grid in the slip path being selected by the slip simulation, that is, calculate the probability of slipping to the target grid each time. The calculation formula is: Among them, P is the influence probability of landslide sliding on the target grid, s i is the number of times the sliding point selects the target grid during the simulation process, m is the number of potential hazards to which the sliding point affecting the target grid belongs, n max is the maximum number of iterations of Monte Carlo random walk simulation, and i and m are positive integers.

4. The method according to claim 1, wherein In step S3, use the analytic hierarchy process to solve the hierarchical relationship weights of the target layer-factor layer and the factor layer-index layer respectively; Use the analytic hierarchy process to solve the weight of a single hierarchical relationship, including: Construct a matrix M, and the elements in the matrix are marked as a ij , combining expert opinions and relevant materials, make pairwise comparisons of the evaluation indicators at two levels to be solved, and fill the determined scale into the corresponding position of the matrix M. The obtained matrix M is the pairwise comparison matrix; among them, a ij represents the importance degree or scale of i compared with j, and the position corresponding to a ij is the i-th row and j-th column in the matrix M; if i = j, it means that i and j are equally important, that is, a ij = 1, and the scales of the main diagonal elements of the matrix M are all 1; a ij > 0 and a ij × a ji = 1; Calculate the weights respectively by using the arithmetic mean method, the geometric mean method, and the eigenvalue method, and calculate the arithmetic mean of the calculation results. Assume the pairwise comparison matrix M is: The weight value w of the i-th factor is obtained by the following three methods i , where First, the weight vector w is obtained by the arithmetic mean method i It is as follows: Second, the weight vector w is obtained by the geometric mean method i It is as follows: Thirdly, for calculating the weight w by the eigenvalue method i , first assume that the largest eigenvalue of the pairwise comparison matrix M is λ max , and its corresponding eigenvector is ξ, then the weight w obtained by the eigenvalue method i is as follows: Conduct a consistency test on the pairwise comparison matrix through the following methods, including: Calculate the eigenvector and the maximum eigenvalue λ of the pairwise comparison matrix max , and the calculation formula is as follows: Among them, M is the pairwise comparison matrix, and w i is the index weight, W = [w1, …, w n T , n is the order of the comparison matrix, and i, j, k, and n are positive integers;​ Calculate the consistency index CI of the pairwise comparison matrix. The calculation formula is: When CI is close to 0, initially judge that the consistency test of the constructed pairwise comparison matrix passes; Calculate the consistency ratio. The calculation formula is: Among them, RI is the average consistency index, which is a parameter value obtained from the lookup table. Usually, when CR < 0.1, it indicates that the consistency test of the pairwise comparison matrix passes.

5. The method according to claim 1, characterized in that In step S4, use the Z-Score method to standardize the values of each index and sum them with weights. The calculation formula is: Among them, R is the risk score value after comprehensively considering risk evaluation indicators, w j is the j-th factor, m is the number of factor layers, w i is the weight of the i-th indicator, n is the number of indicators in the factor layer, x i is the actual observed value or measured value of the i-th indicator, μ is the average value of all observed values, σ is the standard deviation of all indicator data, representing the degree of data dispersion, and i, j, m, n are positive integers.

6. The method according to claim 1, characterized in that In step S4, use the k-means algorithm to classify the landslide hazard risk score values, including: Pre-determine the number of classification categories K, and randomly select K objects {a1, a2, …, a K} as the initial clustering centers of the landslide hazard risk score value data {X1, X2, …, X n}, calculate the distances between each data and each clustering center, allocate them to each clustering center according to the principle of the nearest distance, and recalculate the category centers. Finally, make the sum of the squares of the distances from each data point to the nearest clustering center the smallest. The calculation formula is as follows: Among them, T is the distance objective function of clustering, and q j is the j-th initial clustering center, K is the number of classification categories, and X i is the data to be clustered, n is the number of data, and i, j, and n are positive integers.

7. A landslide hazard grading and risk assessment system considering surface deformation, for implementing the method according to any one of claims 1-6, characterized in that, The system includes: A construction module, which is used to extract the index data corresponding to the disaster-causing factors, disaster-forming factors and disaster-bearing factors, as well as the surface deformation phase and the InSAR surface deformation phase anomaly area based on radar images and terrain data by integrating InSAR technology, deep learning technology and GIS spatial analysis technology, generate a landslide hidden danger risk assessment factor database, and construct a landslide hidden danger risk assessment factor system; wherein, the disaster-causing factors include the surface deformation rate, and the corresponding disaster-causing factor index includes the average surface deformation rate, the disaster-forming factors include slope, elevation and slope unit, and the corresponding disaster-forming factor indexes include the maximum height difference of hidden danger, average slope, and the number of slope units where it is located, and the disaster-bearing factors include buildings, roads and rivers, and the corresponding disaster-bearing factor indexes include the number of threatened buildings, the number of threatened roads and the area of threatened water bodies; A generation module, which is used to determine potential initial slip point grids based on the InSAR surface deformation phase anomaly area, the surface deformation phase, the surface deformation rate, the slope and the elevation, construct a Monte Carlo landslide random walk algorithm for slopes, design a slip truncation condition, conduct Monte Carlo slope slip simulation until the slip truncation condition is reached, count the frequencies of each grid selected in the slip path, convert the frequency data into a spatial vector, generate a potential landslide slip range, which is used as the potential threat range of landslide hidden danger, and is used to extract the index data of hidden danger-related disaster-bearing factors through GIS spatial analysis technology; A solution module, which is used to determine the hierarchical relationship of the landslide hidden danger risk assessment factor system by using the analytic hierarchy process, including the target layer-factor layer and the factor layer-index layer, construct a pairwise comparison matrix, and combine expert opinions and relevant materials to solve the weight coefficients of the two hierarchical relationships, wherein the target layer is the landslide hidden danger risk score value, the factor layer includes disaster-causing factors, disaster-forming factors and disaster-bearing factors, and the index layer includes disaster-causing factor indexes, disaster-forming factor indexes and disaster-bearing factor indexes; An evaluation module, which is used to standardize the values of each index, perform weighted summation on the standardized values according to the weight coefficients of the two hierarchical relationships to obtain the landslide hidden danger risk score value, and classify the landslide hidden danger risk score value to obtain the landslide hidden danger risk level.

8. The system according to claim 7, wherein Extract typical disaster-bearing body targets, that is, disaster-bearing factors, by the following method, and use the deformation phase data and elevation data as the input for model training and inference. Extract the InSAR surface deformation phase anomaly area by the following method: (1) Collect sample data Collect high-resolution remote sensing images of the study area as basic data, select targets with representative features for manual interpretation as reference truth values, combine the marked truth value images and the corresponding original high-resolution images in bands, cut the combined images into several slices of a preset size, and randomly divide them into a sample training set and a sample test set according to a preset ratio; (2) Construct a model Construct a multi-scale feature fusion network model, which adopts an encoder-decoder architecture. In the encoder part, the features of the image are gradually extracted through multiple Depconvgroup modules. Each Depconvgroup module consists of two sets of depthwise separable convolutional layers, a BN normalization layer, and a Relu activation layer. The decoder part is composed of Cbri modules of different scales. Each Cbri module consists of a convolutional layer, a BN normalization layer, and an upsampling layer. After upsampling, the features of different scales are processed into a unified size. Finally, the features of different levels are concatenated in the channel dimension, and a channel attention mechanism module is introduced to improve the model's ability to extract key features; (3) Train the model Use the slice data of the sample training set to train the multi-scale feature fusion network model for multiple rounds. If the error index decreases and tends to be stable, it indicates that the model training is completed. Use the slice data in the sample test set to verify the multi-scale feature fusion network model. If the verification model accuracy does not meet the requirements, adjust the parameters of the model training and retrain until the accuracy requirements are met; (4) Model inference Use the trained multi-scale feature fusion network model to predict all the data in the study area to obtain the distribution results of typical disaster-bearing bodies in the whole area.

9. The system according to claim 7, wherein Construct a slope Monte Carlo landslide random walk algorithm, including: Take the InSAR surface deformation phase anomaly area as the initial potential slip point of the landslide, and construct a random probability calculation formula for the current slip grid to slip to the target grid. The calculation formula is: Among them, P px is the probability that the target grid around the current sliding grid becomes the next sliding grid, n is the number of target grids, β is the slope between the current sliding grid and the target grid, f d and f β are the weight factors for adjusting the sliding direction and the slope respectively. When the direction of the next sliding grid is the same as before, f d is d 2 . When the sliding direction changes by 45°, f d is d. When the sliding direction changes by 90°, f d is 1, f β and d are both initialized input parameters, and f r is the average surface deformation phase rate; Estimate the maximum horizontal slip distance as the slip truncation condition. The calculation formula is: L = K1×ΔH + K2f s + f r + B' (3) where L is the maximum horizontal sliding distance, ΔH is the elevation difference from the top of the landslide to the ground surface, f s and f r are the area of the surface deformation phase aggregation region and the average surface deformation phase rate, respectively, K1 and K2 are correlation coefficients, B' is a constant term, and K1, K2, and B' can be obtained by fitting historical observation data, with the numerical values being 2.54, 0.5, and -50.37, respectively; Statistically calculate the frequency of each grid in the slip path being selected by the slip simulation, that is, calculate the probability of slipping to the target grid each time. The calculation formula is: Among them, P is the influence probability of landslide sliding on the target grid, s i is the number of times the sliding point selects the target grid during the simulation process, m is the number of hidden dangers to which the sliding point affecting the target grid belongs, n max is the maximum number of iterations of the Monte Carlo random walk simulation, and i and m are positive integers.

10. The system according to claim 7, characterized in that, Use the analytic hierarchy process to solve the hierarchical relationship weights of the target layer-factor layer and the factor layer-index layer respectively; Use the analytic hierarchy process to solve the weights of a single hierarchical relationship, including: Construct a matrix M, and the elements in the matrix are marked as a ij , in combination with expert opinions and relevant materials, make pairwise comparisons of the evaluation indicators at two levels to be solved, and fill the determined scale into the corresponding position of the matrix M. The obtained matrix M is the pairwise comparison matrix; among them, a ij represents the importance degree or scale of i compared with j, and the position corresponding to a ij is the i-th row and j-th column in the matrix M; if i = j, it means that i and j are equally important, that is, a ij = 1, and the scales of the main diagonal elements of the matrix M are all 1; a ij > 0 and a ij × a ji = 1; Calculate the weights by using the arithmetic mean method, geometric mean method, and eigenvalue method respectively, and calculate the arithmetic mean of the calculation results. Assume the pairwise comparison matrix M is: The weight value w of the i-th factor is obtained by the following three methods i , where The first one is to obtain the weight vector w by the arithmetic mean method i It is as follows: Second, the weight vector w is obtained by the geometric mean method i It is as follows: Third, for calculating the weight w by the eigenvalue method i , first assume that the maximum eigenvalue of the pairwise comparison matrix M is λ max , and its corresponding eigenvector is ξ. Then the weight w obtained by the eigenvalue method i is as follows: Conduct a consistency test on the pairwise comparison matrix through the following methods, including: Calculate the eigenvector and the maximum eigenvalue λ of the pairwise comparison matrix max , and the calculation formula is as follows: Among them, M is the pairwise comparison matrix, and w i is the index weight, W = [w1, …, w n T , n is the order of the comparison matrix, and i, j, k, and n are positive integers;​ Calculate the consistency index CI of the pairwise comparison matrix. The calculation formula is: When CI is close to 0, it is initially judged that the consistency test of the constructed pairwise comparison matrix passes; Calculate the consistency ratio. The calculation formula is: Among them, RI is the average consistency index, which is a parameter value obtained from the lookup table. Usually, when CR < 0.1, it indicates that the consistency test of the pairwise comparison matrix passes.

Images