Landslide susceptibility assessment methods, systems, equipment and computer-readable storage media

Abstract

A landslide susceptibility assessment method, system, device, and computer-readable storage medium are disclosed, relating to the field of geological hazard risk assessment. Specifically, the method includes determining the deformation rate based on a design matrix and a target vector. The design matrix is ​​a small baseline network matrix composed of time intervals corresponding to interferograms, and the target vector is a column vector corresponding to the unwrapped interferometric phases including slope displacement information. A target feature subset is determined based on disaster-prone environmental characteristic factors and the deformation rate. The target feature subset is input into a preset automatic optimization machine learning model for forward propagation and node splitting calculations to obtain a preliminary probability value for landslides occurring in the slope unit. The landslide susceptibility assessment result is determined based on the preliminary probability value and the deformation rate. This application can improve the accuracy of landslide susceptibility assessment results.

Description

Technical Field

[0001] This application relates to the field of geological hazard risk assessment, specifically to a landslide susceptibility assessment method, system, equipment, and computer-readable storage medium. Background Technology

[0002] Currently, landslide susceptibility assessment is widely used in geological disaster prevention and risk management, which has raised higher demands for improving the accuracy and real-time performance of landslide prediction. Among related technologies, synthetic aperture radar interferometry (InSAR) can identify and quantify surface deformation with high precision, while machine learning models can make probabilistic predictions of landslide susceptibility.

[0003] However, existing technologies have the following problems: InSAR-identified surface deformation data are often only used as a means of post-evaluation verification, making it difficult for models to capture dynamic precursor information of landslides; at the same time, the evaluation results are mostly static probability values, lacking a mechanism to deeply integrate and correct the model prediction results with real-time deformation data, resulting in insufficient accuracy of the evaluation results in reflecting the real-time activity state of the slope.

[0004] Therefore, how to provide a landslide susceptibility assessment method to improve the accuracy of landslide susceptibility assessment results is an urgent problem to be solved. Summary of the Invention

[0005] This application provides a landslide susceptibility assessment method, system, device, and computer-readable storage medium, which can improve the accuracy of landslide susceptibility assessment results.

[0006] In a first aspect, embodiments of this application provide a landslide susceptibility assessment method, the landslide susceptibility assessment method comprising: The deformation rate is determined based on the design matrix and the target vector. The design matrix is ​​a small baseline network matrix composed of time intervals corresponding to the interferogram, and the target vector is a column vector corresponding to the unwrapped interferometric phase, which includes slope displacement information. A subset of target features was determined based on the characteristic factors of the disaster-prone environment and the deformation rate. The target feature subset is input into a preset automatic optimization machine learning model for forward propagation and node splitting calculation to obtain the preliminary probability value of landslide in the slope unit. The landslide susceptibility assessment results were determined based on preliminary probability values ​​and deformation rates.

[0007] In conjunction with the first aspect, in one implementation, determining the deformation rate based on the design matrix and the target vector includes: Substituting the design matrix and the target vector into the following calculation formula yields the deformation rate:

[0008] In the formula, For designing the matrix; It is a column vector composed of M unwrapped interference phases; denoted as the deformation rate.

[0009] In conjunction with the first aspect, in one implementation, the disaster-prone environmental characteristic factors include topographic and geomorphological factors, geological structural factors, hydrological and meteorological factors, and human engineering activity factors. The step of determining the target feature subset based on the disaster-prone environmental characteristic factors and deformation rate includes: For each pair of features among topographic factors, geological structure factors, hydro-meteorological factors, human engineering activity factors, and deformation rate, the correlation coefficient between the two features is calculated. When the correlation coefficient is greater than the preset correlation coefficient threshold, redundant features in the two features are removed, and the first feature subset is obtained based on the features that have not been removed. Each feature in the first feature subset is input into a preset geographic detector to obtain a significance value and an explanatory power value. The significance value is used to evaluate the statistical significance of the influence of each feature, and the explanatory power value is used to quantify the degree of explanatory power of each feature on the spatial distribution of landslides. Features with a significance value greater than a preset significance threshold or an explanatory power value less than a preset explanatory power threshold are removed from the first feature subset to obtain the target feature subset.

[0010] In conjunction with the first aspect, in one implementation, the removal of redundant features from the two features includes: Remove either redundant feature from the two features.

[0011] In conjunction with the first aspect, in one implementation, determining the landslide susceptibility assessment result based on preliminary probability values ​​and deformation rates includes: The landslide susceptibility result is determined based on the preliminary probability value, the preset probability weighting coefficient, the deformation rate, and the preset rate weighting coefficient. The landslide susceptibility assessment zoning map is obtained by classifying the landslide susceptibility results based on the preset natural breakpoint algorithm.

[0012] In conjunction with the first aspect, in one implementation, determining the landslide susceptibility result based on the preliminary probability value, the preset probability weighting coefficient, the deformation rate, and the preset rate weighting coefficient includes: Substituting the preliminary probability value, the preset probability weighting coefficient, the deformation rate, and the preset rate weighting coefficient into the following calculation formula yields the landslide susceptibility result. The calculation formula is as follows:

[0013] In the formula, These are preliminary probability values; Preset probability weighting coefficients; The deformation rate; The preset rate weighting coefficient; This indicates a high susceptibility to landslides.

[0014] In conjunction with the first aspect, in one implementation, the training method for the automatic optimization machine learning model includes: The deformation rate is divided into multiple stability levels based on a pre-defined natural breakpoint algorithm; Non-landslide points were selected as negative samples within the region corresponding to the lowest stability level. The training dataset is determined based on negative and positive samples, wherein the positive samples are known landslide points, and the landslide points include deformation rate and disaster-preparing environmental characteristic factors. The automatic optimization machine learning model is trained based on the training dataset.

[0015] Secondly, embodiments of this application provide a landslide susceptibility assessment system, the landslide susceptibility assessment system comprising: The first processing module is used to determine the deformation rate based on the design matrix and the target vector. The design matrix is ​​a small baseline network matrix composed of time intervals corresponding to the interferogram, and the target vector is a column vector corresponding to the unwrapped interferometric phase including slope displacement information. The second processing module is used to determine the target feature subset based on the disaster-prone environment characteristic factors and deformation rate; The third processing module is used to input the target feature subset into the preset automatic optimization machine learning model for forward propagation and node splitting calculation to obtain the preliminary probability value of landslide in the slope unit. The fourth processing module is used to determine the landslide susceptibility assessment results based on the preliminary probability value and deformation rate.

[0016] Thirdly, embodiments of this application provide a landslide susceptibility assessment device, which includes a processor, a memory, and a landslide susceptibility assessment program stored in the memory and executable by the processor. When the landslide susceptibility assessment program is executed by the processor, it implements the steps of the landslide susceptibility assessment method as described in any of the preceding claims.

[0017] Fourthly, embodiments of this application provide a computer-readable storage medium storing a landslide susceptibility assessment program, wherein when the landslide susceptibility assessment program is executed by a processor, it implements the steps of the landslide susceptibility assessment method as described in any of the preceding claims.

[0018] The beneficial effects of the technical solutions provided in this application include: The deformation rate is determined based on a small baseline network matrix composed of time intervals corresponding to the interferogram and column vectors corresponding to the unwrapped interferometric phases containing slope displacement information. This directly quantifies the dynamic deformation information acquired by InSAR into numerical features usable for modeling, making the deformation rate an input factor rather than just a post-validation tool. A target feature subset is determined based on disaster-prone environmental characteristics and the deformation rate. Real-time monitored dynamic deformation data is integrated with environmental features to form a comprehensive input, improving the model's sensitivity to slope state changes. The target feature subset is input into a pre-set automatic optimization machine learning model for forward propagation and node splitting calculations to obtain a preliminary probability value for landslides in the slope unit, enabling the predicted probability value to respond to dynamic information. Based on the preliminary probability value and deformation rate, the landslide susceptibility assessment result is determined. The susceptibility probability is deeply integrated with real-time deformation data, allowing the final result to reflect the current activity state of the slope and significantly improving the accuracy and reliability of the assessment results. Attached Figure Description

[0019] Figure 1 This is a flowchart illustrating an embodiment of the landslide susceptibility assessment method of this application; Figure 2 This is a schematic diagram illustrating the landslide negative sample selection strategy in the embodiments of the landslide susceptibility assessment method of this application. Figure 2 (a) is to randomly generate negative samples (RS) within the reservoir area. Figure 2 (b) Sampling outside the landslide buffer zone (BBS). Figure 2 (c) Gentle slope-based sampling (GSBS) is performed in a low slope area. Figure 2 (d) Sampling in low-deformation regions (InSAR-based sampling, IBS); Figure 3 This diagram illustrates the landslide susceptibility assessment results of different models in the embodiments of the landslide susceptibility assessment method of this application. Figure 3 (a) shows the landslide susceptibility zoning using the BO-LightGBM model in the IBS-JT-OW setting; Figure 3 (b) The results of the landslide susceptibility assessment of the BO-LightGBM model under the GSBS setting are presented; Figure 3 (c) The distribution of landslide-prone areas obtained by the BO-LightGBM model based on BBS settings; Figure 3 (d) reflects the landslide susceptibility zoning of the BO-LightGBM model under the IBS setting; Figure 3 (e) shows the landslide susceptibility assessment of the BO-LightGBM model under the RS setting; in addition, Figure 3 (f) indicates the landslide susceptibility zoning in the IBS-JT-OW setting using the RF model; Figure 3 (g) is the landslide susceptibility assessment result of the RF model under the IBS setting; Figure 3 (h) shows the distribution of landslide-prone areas in the RF model under the RS setting; Figure 4 This is a schematic diagram of the ROC curves (Receiver Operating Characteristic Curve) for landslide susceptibility assessment of different models in the embodiments of the landslide susceptibility assessment method of this application; Figure 5 This is a schematic diagram of the correlation test of characteristic factors in the landslide susceptibility assessment method of this application; Figure 6 This is a schematic diagram of the hardware structure of the landslide susceptibility assessment device involved in the embodiments of this application. Detailed Implementation

[0020] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present application.

[0021] To make the objectives, technical solutions, and advantages of this application clearer, the embodiments of this application will be described in further detail below with reference to the accompanying drawings.

[0022] Firstly, embodiments of this application provide a method for evaluating landslide susceptibility.

[0023] In one embodiment, reference is made to Figure 1 , Figure 1 This is a flowchart illustrating an embodiment of the landslide susceptibility assessment method of this application. Figure 1 As shown, landslide susceptibility assessment methods include: Step S10: Determine the deformation rate based on the design matrix and the target vector. The design matrix is ​​a small baseline network matrix composed of time intervals corresponding to the interferogram, and the target vector is a column vector corresponding to the unwrapped interferometric phase including slope displacement information.

[0024] As an example, it is assumed that K ascending-orbit SAR images of the same orbit, arranged in a time series, were acquired in the study area, with acquisition times of respectively... The system employs an automatic optimization strategy to select one image as the super master image for registration. In this embodiment, the spatial baseline threshold is preferably set to 5% of the critical baseline, and the temporal baseline threshold is 60 days. Images meeting the above spatiotemporal baseline conditions are multiplied pairwise by conjugate to generate M multi-view difference interferograms. A multi-temporal small baseline interferometric network was thus constructed, and the minimum cost flow (MCF) algorithm was used for phase unwrapping to obtain the unwrapped interferometric phase. Furthermore, a small baseline network design matrix is ​​constructed based on the time intervals corresponding to the interferograms. The design matrix is ​​an M×N dimensional matrix, where N=K 1. Design a matrix to represent the number of time intervals. The expression is as follows:

[0025] in, This refers to the Nth time interval within the time range covered by the Mth interferogram. For time and The k-th unwrapped interferogram generated from two acquired SAR images In, any pixel Interference phase It can be represented as:

[0026] In the formula, The surface deformation phase in the line-of-sight (LOS) direction. For radar wavelength, For cumulative deformations; The phase of the terrain residual; This is the atmospheric delayed phase; For orbital error phase; This is the noise phase.

[0027] Specifically, by establishing a linear observation equation between the design matrix, the deformation rate vector, and the target vector, and then introducing the singular value decomposition method to obtain the generalized pseudo-inverse matrix of the design matrix, the minimum norm solution of the unknown deformation rate is obtained, thereby unifying all unwrapped interferograms under the same reference time series to obtain a high-precision surface deformation rate.

[0028] Step S20: Determine the target feature subset based on the disaster-prone environmental characteristic factors and deformation rate.

[0029] In this exemplary embodiment, the landslide-prone environmental characteristic factor refers to the static or quasi-static geological environmental elements within the study area that affect the development, distribution, and stability of landslides. These elements serve as the basic input variables for landslide susceptibility assessment, characterizing the inherent geological environmental background conditions of the slope unit. Specifically, the static landslide-prone environmental characteristic factor and the dynamic deformation rate are spatially matched and integrated to construct an initial feature set. Correlation analysis is used to calculate the correlation coefficient between each landslide-prone environmental characteristic factor and the deformation rate, and redundant features exceeding a preset correlation coefficient threshold are removed to eliminate multicollinearity. Subsequently, a geographic detector model is used to calculate the significance p-value and explanatory power q-value of the remaining characteristic factors for the spatial distribution of landslides. Low-contribution features with significance values ​​exceeding a preset significance threshold or explanatory power values ​​below a preset explanatory power threshold are removed. Finally, the optimal feature combination that is independent and has strong explanatory power for the spatial distribution of landslides is retained as the target feature subset. This achieves deep fusion of static geological environmental information and dynamic deformation precursor information, reduces input data redundancy, and improves the generalization ability and prediction accuracy of subsequent machine learning models.

[0030] Step S30: Input the target feature subset into the preset automatic optimization machine learning model for forward propagation and node splitting calculation to obtain the preliminary probability value of landslide in the slope unit.

[0031] As an example, the automatic optimization machine learning model in this application embodiment is a composite framework coupled with a LightGBM base classifier and a Bayesian optimization hyperparameter optimization mechanism. The two work together to achieve globally optimal model performance. LightGBM, as the base classifier, utilizes a histogram-based decision tree algorithm to process large-scale, high-dimensional geological environment data. The Bayesian-optimized parameters are automatically optimized to overcome the inefficiencies and susceptibility to local optima inherent in traditional search methods. Specifically, at the parameter setting level, a core parameter optimization space for LightGBM is defined: the search space for the number of leaf nodes (num_leaves) of the decision tree complexity can be preferably set to [10, 100], and the maximum tree depth (max_depth) to [4, 12], ensuring that the number of nodes does not exceed [10, 100] during the optimization process. The algorithm logic constraints are as follows: the number of base learners (n_estimators) can be preferably set to [50, 400], the learning rate (learning_rate) to [0.01, 0.3], and the optimization range of the feature random sampling (colsample_bytree) ratio and the sample random sampling (subsample) ratio to [0.5, 1.0], and the sampling (subsample_freq) frequency is set to activate the sample sampling mechanism. At the same time, the search range of the minimum number of samples in the leaf node (min_child_samples) is introduced to [5, 50] to enhance the model's generalization ability and prevent overfitting.

[0032] It should be understood that in the adaptive optimization process, the initial sampled hyperparameter combinations and their corresponding five-fold cross-validation evaluation indicators (such as AUC values) are first used as the initial observation set. A probabilistic surrogate model between hyperparameters and model performance is constructed using Gaussian process regression. This probabilistic surrogate model predicts the expected performance of unknown parameter combinations through the mean function and outputs the probability distribution of the prediction results through the covariance function. Subsequently, Expected Improvement (EI) is introduced as the acquisition function. In each optimization iteration, by balancing the use of existing optimal regions and the exploration of unknown uncertain regions, the hyperparameter combination that maximizes the expected improvement value is calculated and selected as the input for the next round of model training. As the iteration process continues, the surrogate model continuously updates its posterior distribution based on the newly observed evaluation results, guiding the parameter search space to gradually converge to the global optimum. When the preset maximum number of optimization attempts is reached or the model accuracy tends to stabilize, the system outputs the hyperparameter combination that makes the evaluation index reach the global optimum and solidifies it into the LightGBM base classifier. The finally constructed BO-LightGBM model will be used for subsequent quantitative inference and classification of landslide susceptibility in the entire region.

[0033] It should be noted that, in order to compare and analyze the impact of different negative landslide sample selection strategies on landslide susceptibility assessment, this application employs four different sampling methods, as shown in the attached figures. Figure 2 As shown, Figure 2 (a) is to randomly generate negative samples (RS) within the reservoir area. Figure 2 (b) Sampling outside the landslide buffer zone (BBS). Figure 2 (c) Gentle slope-based sampling (GSBS) is used for sampling in low slope areas. Figure 2(d) Sampling in low-deformation regions (InSAR-based sampling, IBS); at the same time, joint training (JT) and weight overlay (OW) are performed on the basis of IBS sampling, i.e., IBS-JT-OW integration strategy.

[0034] In addition, refer to Figure 3 As shown, to verify the evaluation accuracy and method effectiveness of the BO-LightGBM model, this embodiment simultaneously introduces the Random Forest (RF) algorithm during the model construction stage, and combines it with conventional strategies such as random sampling (RS) and buffer sampling (BBS) to construct multiple benchmark comparison models for quantitative performance comparison analysis; specifically, Figure 3 The paper presents landslide susceptibility zoning and assessment results based on different models and settings, among which... Figure 3 (a) shows the landslide susceptibility zoning using the BO-LightGBM model in the IBS-JT-OW setting; Figure 3 (b) The results of the landslide susceptibility assessment of the BO-LightGBM model under the GSBS setting are presented; Figure 3 (c) The distribution of landslide-prone areas obtained by the BO-LightGBM model based on BBS settings; Figure 3 (d) reflects the landslide susceptibility zoning of the BO-LightGBM model under the IBS setting; Figure 3 (e) shows the landslide susceptibility assessment of the BO-LightGBM model under the RS setting; in addition, Figure 3 (f) indicates the landslide susceptibility zoning in the IBS-JT-OW setting using the RF model; Figure 3 (g) is the landslide susceptibility assessment result of the RF model under the IBS setting; Figure 3 (h) shows the distribution of landslide-prone areas in the RF model under the RS setting.

[0035] It should be understood that, from Figure 3 The model results and the distribution of landslide points show that high and extremely high susceptibility areas (i.e., slope units with the highest susceptibility indices) are mainly concentrated in the Yinmin Town area upstream of the reservoir, the left bank area in the middle reaches, and the area around Dahua Mountain downstream. These areas have steep slopes, fractured rock masses, and are significantly affected by reservoir water fluctuations, which highly matches the deformation-active areas detected by actual surveys and InSAR monitoring, proving the reliability of this method and coupling framework in complex reservoir environments. It should be noted that, referring to... Figure 4As shown, the ROC curve (Receiver Operating Characteristic Curve) can also be used to evaluate the classification performance of the model. The horizontal axis is the false positive rate (FPR), which represents the proportion of correctly predicted non-slope samples; the vertical axis is the true positive rate (TPR), which represents the proportion of correctly predicted slope samples. Both range from 0 to 1. Figure 4 The multiple curves of different colors represent the classification performance of the BO-LightGBM and RF models under different settings. The closer the curve is to the upper left corner, the better the model performance.

[0036] It should be noted that, Figure 4 The different colored lines represent the model performance after optimization using different strategies or dataset combinations. Specifically, the red line represents the BO-LightGBM_IBS-JT-OW model after Bayesian optimization and parameter tuning, with an AUC of 0.962, demonstrating excellent ability to distinguish between positive and negative examples. The black line represents the BO-LightGBM_GSbs model, with an AUC of 0.959, also showing excellent performance, but slightly inferior to the red line model. The pink line represents the RF_IBS-JT-OW random forest model combined with the IBS-JT-OW strategy, with an AUC of 0.952, also exhibiting high performance. The green line... The orange line represents the BO-LightGBM_IBS model, with an AUC of 0.878, indicating good performance. The orange line represents the RF_IBS random forest model incorporating the IBS strategy, with an AUC of 0.852, showing above-average performance. The blue line represents the BO-LightGBM_BBS model, which uses the BBS strategy and employs Bayesian optimization, with an AUC of 0.846. The purple line represents the BO-LightGBM_RS model, with an AUC of 0.816, showing moderate performance. The brown line represents the RF_RS random forest model incorporating the RS strategy, with an AUC of 0.804, indicating relatively low performance among these models.

[0037] Comparative analysis shows that the BO-LightGBM model generally outperforms the RF model under different settings. Among the various settings of the BO-LightGBM model, the IBS-JT-OW setting has the highest AUC value, indicating that the combination of this model and setting has high accuracy and reliability in landslide prediction. In contrast, the RF model has relatively low AUC values ​​under various settings, indicating slightly inferior performance.

[0038] Based on this, the target feature subset is used as a feature vector and input into each decision tree of the automatic optimization machine learning model. For each decision tree, starting from the root node, the split feature identifier and split threshold of the current node are read. The feature value corresponding to the split feature identifier is extracted from the feature vector and compared with the split threshold (the specific value can be determined according to actual needs and is not limited here). Based on the comparison result, the target feature subset is controlled to enter the left child node or the right child node. This comparison and path selection process is recursively executed until the target feature subset reaches the leaf node. The weight value stored in the leaf node is read and the weight values ​​of the leaf nodes output by all decision trees are accumulated to obtain the prediction score. The prediction score is mapped to the (0, 1) interval using a preset probability mapping function to obtain the preliminary probability value of the slope unit causing a landslide. This realizes the nonlinear mapping from the multidimensional feature space to the landslide occurrence probability space. The preset probability mapping function is a mathematical transformation function used to map the original prediction score output by the machine learning model (such as LightGBM) to the probability value in the (0, 1) interval. The Sigmoid function can be preferred.

[0039] Step S40: Determine the landslide susceptibility assessment results based on the preliminary probability value and deformation rate.

[0040] In this exemplary embodiment, the deformation rate is normalized to eliminate dimensional differences and mapped to the same interval as the initial probability value. A weighted fusion is performed using a preset probability weight coefficient and a preset rate weight coefficient. The initial probability value is multiplied by the probability weight coefficient to obtain the static susceptibility contribution value. The normalized deformation rate is multiplied by the rate weight coefficient to obtain the dynamic deformation contribution value. The static susceptibility contribution value and the dynamic deformation contribution value are then linearly weighted and superimposed to obtain the landslide susceptibility assessment result. This achieves a deep fusion of the static susceptibility probability predicted by the machine learning model and the real-time deformation dynamic information monitored by InSAR. The assessment result includes both the inherent susceptibility determined by the geological environment and the current activity state of the slope, significantly improving the accuracy and timeliness of the landslide susceptibility assessment.

[0041] This application determines the deformation rate based on a small baseline network matrix composed of time intervals corresponding to interferograms and column vectors corresponding to unwrapped interferometric phases including slope displacement information. It directly quantifies the dynamic deformation information acquired by InSAR into numerical features usable for modeling, making the deformation rate an input factor rather than just a post-validation factor. A target feature subset is determined based on disaster-prone environmental characteristics and the deformation rate. Real-time monitored dynamic deformation data is integrated with environmental features to form a comprehensive input, improving the model's sensitivity to slope state changes. The target feature subset is input into a pre-defined automatic optimization machine learning model for forward propagation and node splitting calculations to obtain a preliminary probability value for landslide occurrence in the slope unit, enabling the predicted probability value to respond to dynamic information. Based on the preliminary probability value and deformation rate, the landslide susceptibility assessment result is determined. The susceptibility probability is deeply integrated with real-time deformation data, allowing the final result to reflect the current activity state of the slope and significantly improving the accuracy and reliability of the assessment results.

[0042] Further, in one embodiment, determining the deformation rate based on the design matrix and the target vector includes: Substituting the design matrix and the target vector into the following calculation formula yields the deformation rate:

[0043] In the formula, For designing the matrix; It is a column vector composed of M unwrapped interference phases; denoted as the deformation rate.

[0044] As an example, in this embodiment of the application, to avoid the rank deficiency problem of the equation system caused by directly solving the cumulative deformation, the parameter to be solved (cumulative deformation) is converted into the average deformation rate of adjacent time periods; let the deformation rate vector between two adjacent images be... Extending this to all M interferograms, we construct a system of linear matrix equations concerning the deformation rate:

[0045] in, for The small baseline network design matrix (composed of time intervals), It is a column vector composed of the M unwrapped interference phases.

[0046] It should be noted that the set spatiotemporal baseline threshold causes the image network to break into multiple isolated small baseline subsets, which makes the design matrix... The matrix exhibits singularity; therefore, singular value decomposition (SVD) is introduced to obtain the matrix. The generalized pseudoinverse matrix, decomposed into matrix. ( V is the left singular matrix obtained after performing singular value decomposition on the design matrix A, and V is the right singular matrix obtained after performing singular value decomposition on the design matrix A. (Given its transpose, both being orthogonal matrices), the deformation rate can be obtained. The minimum norm solution is:

[0047] By using SVD to solve the problem, the rank deficiency caused by isolated subsets is effectively eliminated, and all unwrapped interferograms are unified under the same reference time series.

[0048] Further, in one embodiment, the disaster-prone environmental characteristic factors include topographic and geomorphological factors, geological structure factors, hydrological and meteorological factors, and human engineering activity factors. The step of determining the target feature subset based on the disaster-prone environmental characteristic factors and deformation rate includes: For each pair of features among topographic factors, geological structure factors, hydro-meteorological factors, human engineering activity factors, and deformation rate, the correlation coefficient between the two features is calculated. When the correlation coefficient is greater than the preset correlation coefficient threshold, redundant features in the two features are removed, and the first feature subset is obtained based on the features that have not been removed. Each feature in the first feature subset is input into a preset geographic detector to obtain a significance value and an explanatory power value. The significance value is used to evaluate the statistical significance of the influence of each feature, and the explanatory power value is used to quantify the degree of explanatory power of each feature on the spatial distribution of landslides. Features with a significance value greater than a preset significance threshold or an explanatory power value less than a preset explanatory power threshold are removed from the first feature subset to obtain the target feature subset.

[0049] As an example, this application initially selects disaster-prone environmental characteristic factors covering topography, geological structure, hydrology and meteorology, and human activities, specifically including: elevation, slope, aspect, plane curvature, profile curvature, topographic relief, flow intensity index (SPI), topographic humidity index (TWI), distance from fault, distance from dam site, distance from road, distance from dead water level, normalized difference vegetation index (NDVI), normalized difference building index (NDBI), multi-year average rainfall, POI (point of interest) kernel density, stratigraphic lithology, landform type, land use, etc. Specifically, for the initial feature set composed of topography factors, geological structure factors, hydrology and meteorology factors, human engineering activity factors, and deformation rate, the Pearson correlation coefficient r between any two feature factors in the set is calculated to quantify the degree of linear correlation between features. The calculated correlation coefficient is compared with a preset correlation coefficient threshold (the specific value can be determined according to actual needs and is not limited here; for example, the preset correlation coefficient threshold can preferably be 0.7).

[0050] It should be noted that when the correlation coefficient between any two feature factors exceeds a preset correlation coefficient threshold, it is determined that these two feature factors exhibit multicollinearity redundancy. One of these feature factors is removed to eliminate data redundancy, and the remaining feature factors are used to construct the first feature subset. This reduces the dimensionality of the input data and avoids the risk of overfitting in subsequent machine learning models due to high correlation between features, thereby improving the stability and convergence speed of model training. (Refer to...) Figure 5 As shown, the correlation coefficients of feature b and feature f are both 0.92 and 0.92 > 0.7, so either feature can be removed to eliminate data redundancy.

[0051] Based on this, each feature factor in the first feature subset is input into the geographic detector model. The explanatory power of each feature factor on the spatial distribution of landslides is quantified by calculating the q-value. The q-value ranges from 0 to 1. At the same time, the significance test results are combined for comprehensive evaluation. Feature factors that fail the significance test, i.e., the p-value is greater than the preset explanatory power threshold (the specific value can be determined according to actual needs and is not limited here. For example, the preset correlation coefficient threshold can be preferably 0.05) or the q-value is extremely low, are eliminated. The remaining independent feature factors that have strong explanatory power on the spatial distribution of landslides are retained to form the target feature subset. This ensures that the feature factors input into the model not only eliminate multicollinearity but also have significant geological interpretation significance, thereby improving the feature selection quality and generalization ability of the model.

[0052] Furthermore, in one embodiment, the removal of redundant features from the two features includes: Remove either redundant feature from the two features.

[0053] In an exemplary embodiment of this application, when the correlation coefficient between any two feature factors is detected to be greater than a preset correlation coefficient threshold, it is determined that the two feature factors have multicollinearity redundancy with overlapping information, and a redundant feature removal operation is performed; that is, one of the two highly correlated feature factors is selected for removal and the other is retained, or the strength of the geological indication significance of the two is judged based on experience and the weaker feature is removed. By utilizing the characteristic that highly correlated features carry highly similar information, the number of input variables is reduced while retaining the main geological environment information, thereby eliminating the high correlation between features, reducing the dimensionality of the input data and avoiding the risk of overfitting caused by the high correlation between features in the subsequent machine learning model, improving the stability and convergence speed of model training, and finally obtaining the first feature subset based on the features that have not been removed.

[0054] Furthermore, in one embodiment, determining the landslide susceptibility assessment result based on the preliminary probability value and deformation rate includes: The landslide susceptibility result is determined based on the preliminary probability value, the preset probability weighting coefficient, the deformation rate, and the preset rate weighting coefficient. The landslide susceptibility assessment zoning map is obtained by classifying the landslide susceptibility results based on the preset natural breakpoint algorithm.

[0055] As an example, in the embodiments of this application, the specific values ​​of the preset probability weight coefficient and the preset rate weight coefficient can be determined according to actual needs, as long as the sum of the two is 1, and there is no limitation here. For example, the preset probability weight coefficient can preferably be 0.61, and the preset rate weight coefficient can preferably be 0.39. The preset natural breakpoint algorithm is a classification optimization method based on data distribution characteristics. Its core principle is to minimize the variance within the same level and maximize the variance between levels after classification through iterative calculation, thereby automatically dividing continuous data into multiple levels with significant differences, so that the data difference within the same level is minimized and the data difference between different levels is maximized. It is used to objectively identify the natural clustering characteristics in the data without relying on a subjectively set fixed threshold.

[0056] Specifically, the deformation rate is normalized to eliminate dimensional differences and map the value to the same interval as the preliminary probability value. The normalized deformation rate is multiplied by a preset rate weighting coefficient to obtain the dynamic deformation contribution value. The preliminary probability value is multiplied by a preset probability weighting coefficient to obtain the static susceptibility contribution value. The dynamic deformation contribution value and the static susceptibility contribution value are linearly weighted and superimposed to obtain the landslide susceptibility result. This achieves a deep fusion of the static susceptibility probability predicted by the machine learning model and the real-time deformation dynamic information monitored by InSAR. The evaluation result includes both the inherent susceptibility determined by the geological environment and the current activity state of the slope, significantly improving the accuracy and timeliness of landslide susceptibility evaluation.

[0057] It should be noted that the natural discontinuity method is used in the geographic information system to classify the continuous landslide susceptibility index. This method is based on the inherent statistical distribution characteristics of the data and iteratively calculates the troughs of the data distribution as discontinuities, so that the variance within each level is minimized while the variance between different levels is maximized. Based on this mathematical principle, the continuous susceptibility index is automatically discretized into five susceptibility levels: extremely low, low, medium, high, and extremely high. Finally, a landslide susceptibility evaluation zoning map reflecting the spatial distribution differences of landslide risk is output, realizing the objective classification and visualization of the evaluation results.

[0058] Further, in one embodiment, determining the landslide susceptibility result based on the preliminary probability value, the preset probability weighting coefficient, the deformation rate, and the preset rate weighting coefficient includes: Substituting the preliminary probability value, the preset probability weighting coefficient, the deformation rate, and the preset rate weighting coefficient into the following calculation formula yields the landslide susceptibility result. The calculation formula is as follows:

[0059] In the formula, These are preliminary probability values; Preset probability weighting coefficients; The deformation rate; The preset rate weighting coefficient; This indicates a high susceptibility to landslides.

[0060] As an example, in the embodiments of this application, the preliminary probability value is... Preset probability weighting coefficients Deformation rate and preset rate weighting coefficient Substituting into the following calculation formula yields the landslide susceptibility result. :

[0061] Furthermore, in one embodiment, the training method for the automatic optimization machine learning model includes: The deformation rate is divided into multiple stability levels based on a pre-defined natural breakpoint algorithm; Non-landslide points were selected as negative samples within the region corresponding to the lowest stability level. The training dataset is determined based on negative and positive samples, wherein the positive samples are known landslide points, and the landslide points include deformation rate and disaster-preparing environmental characteristic factors. The automatic optimization machine learning model is trained based on the training dataset.

[0062] As an example, the surface deformation rate of the study area was obtained using SBAS-InSAR technology, and Kriging spatial interpolation was used to fill in the gaps caused by local incoherence. The Jenks method, based on the statistical principle of minimum variance within the same level and maximum variance between levels, was used to automatically classify the absolute values ​​of the deformation rate of the entire area into four levels: extremely weak, weak, relatively strong, and strong (preferably [0~4, 4~10, 10~18, 18~76 mm / yr] in this embodiment). Non-slope points were selected as negative samples within the area corresponding to the lowest stability level (e.g., 0~4 mm / yr). In this process, an equal number of non-landslide negative samples are randomly generated within the lowest level of deformation, equal to the number of positive samples. A training dataset is determined based on the negative and positive samples, where the positive samples are known landslide points. The sample features (i.e., landslide point features) include deformation rate and disaster-prone environmental factors. The extracted deformation rate is treated as an independent feature column and concatenated with static environmental factors such as elevation and slope to form the training dataset. An automatic optimization machine learning model is trained using this training dataset. The model treats the deformation rate factor equally with other factors and participates in the decision tree splitting calculation, thereby establishing a mapping relationship between deformation rate and the probability of landslide occurrence. It should be noted that the above-described classification of the absolute value of deformation rate is only a presentation of an example; more or fewer classifications can be implemented according to actual needs.

[0063] Secondly, embodiments of this application also provide a landslide susceptibility assessment system, which includes: The first processing module is used to determine the deformation rate based on the design matrix and the target vector. The design matrix is ​​a small baseline network matrix composed of time intervals corresponding to the interferogram, and the target vector is a column vector corresponding to the unwrapped interferometric phase including slope displacement information. The second processing module is used to determine the target feature subset based on the disaster-prone environment characteristic factors and deformation rate; The third processing module is used to input the target feature subset into the preset automatic optimization machine learning model for forward propagation and node splitting calculation to obtain the preliminary probability value of landslide in the slope unit. The fourth processing module is used to determine the landslide susceptibility assessment results based on the preliminary probability value and deformation rate.

[0064] Furthermore, in one embodiment, the first processing module is specifically used for: Substituting the design matrix and the target vector into the following calculation formula yields the deformation rate:

[0065] In the formula, For designing the matrix; It is a column vector composed of M unwrapped interference phases; denoted as the deformation rate.

[0066] Furthermore, in one embodiment, the disaster-prone environmental characteristic factors include topographic factors, geological structure factors, hydrological and meteorological factors, and human engineering activity factors, and the second processing module is specifically used for: For each pair of features among topographic factors, geological structure factors, hydro-meteorological factors, human engineering activity factors, and deformation rate, the correlation coefficient between the two features is calculated. When the correlation coefficient is greater than the preset correlation coefficient threshold, redundant features in the two features are removed, and the first feature subset is obtained based on the features that have not been removed. Each feature in the first feature subset is input into a preset geographic detector to obtain a significance value and an explanatory power value. The significance value is used to evaluate the statistical significance of the influence of each feature, and the explanatory power value is used to quantify the degree of explanatory power of each feature on the spatial distribution of landslides. Features with a significance value greater than a preset significance threshold or an explanatory power value less than a preset explanatory power threshold are removed from the first feature subset to obtain the target feature subset.

[0067] Furthermore, in one embodiment, the second processing module is specifically used for: Remove either redundant feature from the two features.

[0068] Furthermore, in one embodiment, the fourth processing module is specifically used for: The landslide susceptibility result is determined based on the preliminary probability value, the preset probability weighting coefficient, the deformation rate, and the preset rate weighting coefficient. The landslide susceptibility assessment zoning map is obtained by classifying the landslide susceptibility results based on the preset natural breakpoint algorithm.

[0069] Furthermore, in one embodiment, the fourth processing module is specifically used for: Substituting the preliminary probability value, the preset probability weighting coefficient, the deformation rate, and the preset rate weighting coefficient into the following calculation formula yields the landslide susceptibility result. The calculation formula is as follows:

[0070] In the formula, These are preliminary probability values; Preset probability weighting coefficients; The deformation rate; The preset rate weighting coefficient; This indicates a high susceptibility to landslides.

[0071] Furthermore, in one embodiment, the third processing module is specifically used for: The deformation rate is divided into multiple stability levels based on a pre-defined natural breakpoint algorithm; Non-landslide points were selected as negative samples within the region corresponding to the lowest stability level. The training dataset is determined based on negative and positive samples, wherein the positive samples are known landslide points, and the landslide points include deformation rate and disaster-preparing environmental characteristic factors. The automatic optimization machine learning model is trained based on the training dataset.

[0072] This application determines the deformation rate based on a small baseline network matrix composed of time intervals corresponding to interferograms and column vectors corresponding to unwrapped interferometric phases including slope displacement information. It directly quantifies the dynamic deformation information acquired by InSAR into numerical features usable for modeling, making the deformation rate an input factor rather than just a post-validation factor. A target feature subset is determined based on disaster-prone environmental characteristics and the deformation rate. Real-time monitored dynamic deformation data is integrated with environmental features to form a comprehensive input, improving the model's sensitivity to slope state changes. The target feature subset is input into a pre-defined automatic optimization machine learning model for forward propagation and node splitting calculations to obtain a preliminary probability value for landslide occurrence in the slope unit, enabling the predicted probability value to respond to dynamic information. Based on the preliminary probability value and deformation rate, the landslide susceptibility assessment result is determined. The susceptibility probability is deeply integrated with real-time deformation data, allowing the final result to reflect the current activity state of the slope and significantly improving the accuracy and reliability of the assessment results.

[0073] The functions of each module in the landslide susceptibility assessment system correspond to the steps in the landslide susceptibility assessment method embodiment, and their functions and implementation processes will not be described in detail here.

[0074] Thirdly, this application provides a landslide susceptibility assessment device, which can be a personal computer (PC), laptop computer, server, or other device with data processing capabilities.

[0075] Reference Figure 6 , Figure 6 This is a schematic diagram of the hardware structure of the landslide susceptibility assessment device involved in the embodiments of this application. In the embodiments of this application, the landslide susceptibility assessment device may include a processor, a memory, a communication interface, and a communication bus.

[0076] The communication bus can be of any type and is used to interconnect the processor, memory, and communication interface.

[0077] The communication interface includes input / output (I / O) interfaces, physical interfaces, and logical interfaces used for interconnecting devices within the landslide susceptibility assessment equipment, as well as interfaces used for interconnecting the landslide susceptibility assessment equipment with other devices (such as other computing devices or user equipment). Physical interfaces can be Ethernet interfaces, fiber optic interfaces, ATM interfaces, etc.; user equipment can be displays, keyboards, etc.

[0078] Memory can be various types of storage media, such as random access memory (RAM), read-only memory (ROM), non-volatile RAM (NVRAM), flash memory, optical storage, hard disk, programmable ROM (PROM), erasable PROM (EPROM), electrically erasable PROM (EEPROM), etc.

[0079] The processor can be a general-purpose processor, which can call the landslide susceptibility assessment program stored in the memory and execute the landslide susceptibility assessment method provided in the embodiments of this application. For example, the general-purpose processor can be a central processing unit (CPU). The method executed when the landslide susceptibility assessment program is called can be referred to in the various embodiments of the landslide susceptibility assessment method of this application, and will not be repeated here.

[0080] Those skilled in the art will understand that Figure 6 The hardware structure shown does not constitute a limitation of this application and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0081] Fourthly, embodiments of this application also provide a readable storage medium.

[0082] This application has a readable storage medium storing a landslide susceptibility assessment program, wherein when the landslide susceptibility assessment program is executed by a processor, it implements the steps of the landslide susceptibility assessment method as described above.

[0083] The method implemented when the landslide susceptibility assessment procedure is executed can be referred to in the various embodiments of the landslide susceptibility assessment method of this application, and will not be repeated here.

[0084] The terms "comprising" and "having," and any variations thereof, in the specification, claims, and accompanying drawings of this application are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to such process, method, product, or apparatus. The terms "first," "second," and "third," etc., are used to distinguish different objects, etc., and do not indicate a sequence, nor do they limit "first," "second," and "third" to different types.

[0085] In the description of the embodiments of this application, terms such as "exemplary," "for example," or "for instance" are used to indicate examples, illustrations, or explanations. Any embodiment or design described as "exemplary," "for example," or "for instance" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of terms such as "exemplary," "for example," or "for instance" is intended to present the relevant concepts in a concrete manner.

[0086] In the description of the embodiments of this application, unless otherwise stated, " / " means "or". For example, A / B can mean A or B. The "and / or" in the text is merely a description of the relationship between related objects, indicating that there can be three relationships. For example, A and / or B can mean: A exists alone, A and B exist simultaneously, and B exists alone. In addition, in the description of the embodiments of this application, "multiple" means two or more.

[0087] In some processes described in the embodiments of this application, multiple operations or steps are included in a specific order. However, it should be understood that these operations or steps may not be executed in the order they appear in the embodiments of this application, or they may be executed in parallel. The sequence number of the operation is only used to distinguish different operations, and the sequence number itself does not represent any execution order. In addition, these processes may include more or fewer operations, and these operations or steps may be executed sequentially or in parallel, and these operations or steps may be combined.

[0088] It should be noted that the sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0089] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes several instructions to cause a terminal device to execute the methods described in the various embodiments of this application.

[0090] The above are merely preferred embodiments of this application and do not limit the patent scope of this application. Any equivalent structural or procedural transformations made using the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

Claims

1. A method for evaluating landslide susceptibility, characterized in that, The landslide susceptibility assessment method includes: The deformation rate is determined based on the design matrix and the target vector. The design matrix is ​​a small baseline network matrix composed of time intervals corresponding to the interferogram, and the target vector is a column vector corresponding to the unwrapped interferometric phase, which includes slope displacement information. For each pair of features among topographic factors, geological structure factors, hydro-meteorological factors, human engineering activity factors, and deformation rate, the correlation coefficient between the two features is calculated. When the correlation coefficient is greater than the preset correlation coefficient threshold, redundant features in the two features are removed, and the first feature subset is obtained based on the features that have not been removed. Each feature in the first feature subset is input into a preset geographic detector to obtain a significance value and an explanatory power value. The significance value is used to evaluate the statistical significance of the influence of each feature, and the explanatory power value is used to quantify the degree of explanatory power of each feature on the spatial distribution of landslides. Features with a significance value greater than a preset significance threshold or an explanatory power value less than a preset explanatory power threshold are removed from the first feature subset to obtain the target feature subset; The target feature subset is input into a preset automatic optimization machine learning model for forward propagation and node splitting calculation to obtain the preliminary probability value of landslide in the slope unit. The training method for the automatic optimization machine learning model includes: The deformation rate is divided into multiple stability levels based on a pre-defined natural breakpoint algorithm; Non-landslide points were selected as negative samples within the region corresponding to the lowest stability level. The training dataset is determined based on negative and positive samples, wherein the positive samples are known landslide points, and the landslide points include deformation rate and disaster-preparing environmental characteristic factors. The automatic optimization machine learning model is trained based on the training dataset; The landslide susceptibility assessment results were determined based on preliminary probability values ​​and deformation rates.

2. The landslide susceptibility assessment method as described in claim 1, characterized in that, The determination of deformation rate based on the design matrix and target vector includes: Substituting the design matrix and the target vector into the following calculation formula yields the deformation rate: In the formula, For designing the matrix; It is a column vector composed of M unwrapped interference phases; denoted as the deformation rate.

3. The landslide susceptibility assessment method as described in claim 1, characterized in that, The process of removing redundant features from the two features includes: Remove either redundant feature from the two features.

4. The landslide susceptibility assessment method as described in claim 1, characterized in that, The landslide susceptibility assessment results determined based on preliminary probability values ​​and deformation rates include: The landslide susceptibility result is determined based on the preliminary probability value, the preset probability weighting coefficient, the deformation rate, and the preset rate weighting coefficient. The landslide susceptibility assessment zoning map is obtained by classifying the landslide susceptibility results based on the preset natural breakpoint algorithm.

5. The landslide susceptibility assessment method as described in claim 4, characterized in that, The determination of landslide susceptibility based on preliminary probability values, preset probability weighting coefficients, deformation rates, and preset rate weighting coefficients includes: Substituting the preliminary probability value, the preset probability weighting coefficient, the deformation rate, and the preset rate weighting coefficient into the following calculation formula yields the landslide susceptibility result. The calculation formula is as follows: In the formula, These are preliminary probability values; Preset probability weighting coefficients; The deformation rate; The preset rate weighting coefficient; This indicates a high susceptibility to landslides.

6. A landslide susceptibility assessment system, characterized in that, The landslide susceptibility assessment system includes: The first processing module is used to determine the deformation rate based on the design matrix and the target vector. The design matrix is ​​a small baseline network matrix composed of time intervals corresponding to the interferogram, and the target vector is a column vector corresponding to the unwrapped interferometric phase including slope displacement information. The second processing module is used to calculate the correlation coefficient between two features for each pair of features in topographic and geomorphological factors, geological structure factors, hydro-meteorological factors, human engineering activity factors and deformation rate. When the correlation coefficient is greater than the preset correlation coefficient threshold, redundant features in the two features are removed, and the first feature subset is obtained based on the features that are not removed. Each feature in the first feature subset is input into a preset geographic detector to obtain a significance value and an explanatory power value. The significance value is used to evaluate the statistical significance of the influence of each feature, and the explanatory power value is used to quantify the degree of explanatory power of each feature on the spatial distribution of landslides. Features with a significance value greater than a preset significance threshold or an explanatory power value less than a preset explanatory power threshold are removed from the first feature subset to obtain the target feature subset; The third processing module is used to input the target feature subset into the preset automatic optimization machine learning model for forward propagation and node splitting calculation to obtain the preliminary probability value of landslide in the slope unit. The deformation rate is divided into multiple stability levels based on a pre-defined natural breakpoint algorithm; Non-landslide points were selected as negative samples within the region corresponding to the lowest stability level. The training dataset is determined based on negative and positive samples, wherein the positive samples are known landslide points, and the landslide points include deformation rate and disaster-preparing environmental characteristic factors. The automatic optimization machine learning model is trained based on the training dataset; The fourth processing module is used to determine the landslide susceptibility assessment results based on the preliminary probability value and deformation rate.

7. A landslide susceptibility assessment device, characterized in that, The landslide susceptibility assessment device includes a processor, a memory, and a landslide susceptibility assessment program stored in the memory and executable by the processor, wherein when the landslide susceptibility assessment program is executed by the processor, it implements the steps of the landslide susceptibility assessment method as described in any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a landslide susceptibility assessment program, wherein when the landslide susceptibility assessment program is executed by a processor, it implements the steps of the landslide susceptibility assessment method as described in any one of claims 1 to 5.

Images