A landslide susceptibility evaluation method based on category-based feature enhancement

Abstract

This invention provides a landslide susceptibility assessment method based on categorical feature enhancement, belonging to the field of landslide prediction. The method includes: preprocessing multi-class data of the landslide disaster research area, deriving the original values ​​of evaluation factors corresponding to landslide points to obtain an original dataset; training a categorical feature enhancement algorithm and performing grid search to obtain a categorical feature enhancement algorithm with optimal parameters; using an adaptive enhancement model as a meta-learner, and using the categorical feature enhancement algorithm with optimal parameters as a base learner for adaptive enhancement ensemble modeling to obtain an ensemble model; substituting the training set into the ensemble model for training simulation, and importing grid points into the trained ensemble model for landslide susceptibility assessment, outputting a landslide susceptibility prediction map. This invention solves the problems of low adaptability of existing machine learning models to complex data and low accuracy in susceptibility assessment.

Description

Technical Field

[0001] This invention belongs to the field of landslide disaster prediction, and in particular relates to a landslide susceptibility assessment method based on categorical feature enhancement. Background Technology

[0002] With the increasing frequency of landslides and their significant impact on life and property, accurate landslide susceptibility assessments are of paramount importance. Traditional machine learning methods often suffer from insufficient capture of nonlinear features when dealing with landslide prediction. Therefore, integrating multiple models has become an effective means to improve prediction accuracy.

[0003] Single machine learning models may encounter limitations when dealing with complex problems, such as susceptibility to overfitting, sensitivity to outliers, and insufficient generalization ability. Combining different machine learning models can fully leverage their respective strengths, such as the interpretability of decision trees and the boundary optimization capabilities of support vector machines. To overcome these shortcomings, researchers often use ensemble methods that combine multiple machine learning models with search algorithms and optimization strategy frameworks to improve model performance. Summary of the Invention

[0004] In view of the above-mentioned shortcomings in the existing technology, the present invention provides a landslide susceptibility assessment method based on categorical feature enhancement, which solves the problems of low adaptability of existing different machine learning models to complex data and low accuracy of susceptibility assessment.

[0005] To achieve the above objectives, the technical solution adopted by this invention is as follows: Firstly, this invention provides a landslide susceptibility assessment method based on categorical feature enhancement, comprising the following steps:

[0006] S1. Preprocess the various types of data obtained from the landslide disaster study area, and export the original values ​​of all initial evaluation factors corresponding to all landslide points in the landslide disaster study area to obtain the original dataset.

[0007] S2. Establish a categorical feature enhancement model using a categorical feature enhancement algorithm, and train the categorical feature enhancement model using the original dataset. Use the area under the curve as the training metric, and use a grid search algorithm to search for the optimal parameters of the categorical feature enhancement model to obtain a categorical feature enhancement model with optimal parameters.

[0008] S3. Using the adaptive enhancement model and combining it with the categorical feature enhancement model with optimal parameters, we obtain the adaptive enhancement-categorical feature enhancement ensemble model.

[0009] S4. Process the original dataset to obtain an optimized dataset, and use the optimized dataset to train and simulate the adaptive augmentation-categorical feature enhancement ensemble model to obtain the trained adaptive augmentation-categorical feature enhancement ensemble model. Use evaluation metrics to test the accuracy of the trained adaptive augmentation-categorical feature enhancement ensemble model.

[0010] S5. Based on the trained adaptive enhancement-categorical feature enhancement ensemble model, use the categorical feature enhancement classifier to generate multiple decision trees and predict the grid points in the landslide disaster research area. Then, use the adaptive enhancement classifier to summarize the predictions and output the susceptibility probability value corresponding to each grid cell.

[0011] S6. Use the natural breakpoint method to process the susceptibility probability value, obtain the landslide susceptibility evaluation result, and output the landslide susceptibility prediction map.

[0012] The beneficial effects of this invention are as follows: This invention acquires multi-source data of the landslide disaster research area, processes the raw data, and uses a software multi-value extraction tool to extract the raw data table of all evaluation factors of the sample set according to the landslide point data, and packages it to construct the raw dataset; it uses a grid search algorithm to optimize the parameters of the categorical feature enhancement model, and then performs adaptive enhancement ensemble on the optimized categorical feature enhancement model, which improves the accuracy of the adaptive enhancement-categorical feature enhancement ensemble model; it then substitutes the sample set into the adaptive enhancement-categorical feature enhancement ensemble model for training, providing an efficient adaptive enhancement-categorical feature enhancement ensemble model; finally, it imports the values ​​of all grid points into the trained model for landslide disaster susceptibility analysis, classifies the research area into levels, and outputs a landslide disaster susceptibility prediction map, thereby improving the performance indicators of landslide susceptibility evaluation.

[0013] Further, S1 includes the following steps:

[0014] S101. Obtain various types of data for the landslide disaster research area;

[0015] S102. According to the preset sample ratio of landslide points and non-landslide points, randomly generate non-landslide points outside the preset diameter of landslide points;

[0016] S103. Unify the spatial resolution and coordinate system of multiple types of data, and transform the multiple types of data after unifying the coordinate system into layers of the same raster size through resampling to obtain the transformed raster data.

[0017] S104. Based on the converted raster data, use the software to extract the point tool, and according to the landslide point data, extract the original data table of all initial evaluation factors, export the original values ​​of all initial evaluation factors, and obtain the original dataset.

[0018] Furthermore, step S103 includes the following steps:

[0019] S1031. Unify the spatial resolution and coordinate system of multiple types of data;

[0020] S1032. Resample the multi-class data after unifying the coordinate system, select the elevation raster as the reference image, set the input raster image to the raster image to be resampled, set the output raster dataset to the newly created raster, set the output pixels to the same elevation raster as the layer, and set the resampling to the nearest, to obtain the converted raster data.

[0021] The beneficial effects of the above-mentioned further solutions are as follows: the present invention performs raster processing on the acquired multi-type data to obtain the raster with relevant settings, which improves the effectiveness and accuracy of the acquired data and increases the processing speed of the enhanced model on the data.

[0022] Furthermore, S2 includes the following steps:

[0023] S201. Establish a categorical feature enhancement model by combining the categorical feature enhancement algorithm, and train the categorical feature enhancement model using the original dataset. Perform grid search using the grid search class in the scikit-learn library.

[0024] S202. Using the fit method, the area under the curve is used as the training metric to search for the optimal parameters of the categorical feature enhancement model, and the best parameter combination is obtained. Based on the best parameter combination, the categorical feature enhancement model with the best parameters is obtained.

[0025] Furthermore, step S201 includes the following steps:

[0026] S2011. Create a base learner containing instances of the categorical feature enhancement model and define a parameter grid containing the maximum number of iterations and candidate learning rate values.

[0027] S2012. Create a grid search instance, passing in the categorical feature boosting classifier, parameter grid, number of folds for cross-validation, scoring criteria, and parameters for parallel processing, and use the grid search class to perform a grid search on the passed parameters.

[0028] The beneficial effects of the above-mentioned further solutions are as follows: This invention utilizes the fit method and uses the area under the curve as a training metric, thereby improving the efficiency of machine learning models in finding the optimal parameter combination.

[0029] Furthermore, the specific steps of S3 are as follows:

[0030] By using the adaptive augmentation classifier in the scikit-learn library to call the adaptive augmentation model, the adaptive augmentation model is used as the meta-learner, and the categorical feature enhancement model with the optimal parameters is used as the base learner. The meta-learner and the base learner are then integrated through adaptive augmentation to obtain the adaptive augmentation-categorical feature enhancement ensemble model.

[0031] Furthermore, the specific steps of S4 are as follows:

[0032] S401. Process the original dataset by setting the evaluation factor values ​​(excluding ID and attribute data Y) to X and setting the attribute data Y to the training target Y to obtain the optimized dataset.

[0033] S402. Randomly shuffle the optimized dataset according to the preset set ratio and divide it into a training set and a test set. Substitute the training set into the adaptive enhancement-categorical feature enhancement ensemble model for training simulation to obtain the trained adaptive enhancement-categorical feature enhancement ensemble model.

[0034] S403. Plot the receiver operating characteristic curve using the test set, obtain the area under the curve value, and import the area under the curve function code into the adaptive augmentation-categorical feature enhancement ensemble model. Use the area under the curve value to test the accuracy of the trained adaptive augmentation-categorical feature enhancement ensemble model.

[0035] The beneficial effects of the above-mentioned further solutions are as follows: This invention uses an adaptive enhancement model as a meta-learner and the CatBoost algorithm as a base learner. The integration method of multiple machine learning models enhances the model's adaptability to complex data and improves the predictive performance of the machine learning model.

[0036] Furthermore, the specific steps of S5 are as follows:

[0037] S501. Based on the trained adaptive enhancement-categorical feature enhancement ensemble model, the input layer receives the evaluation factors of the study area, and the adaptive enhancement classifier combines the base learners of multiple categorical feature enhancement classifiers.

[0038] S502. Based on the optimal parameters, use categorical features to boost the classifier to generate multiple decision trees, and predict the grid points in the landslide disaster study area based on the evaluation factors of the study area.

[0039] S503. Use an adaptive augmentation classifier to summarize the prediction results of the decision tree, output the susceptibility probability value corresponding to each grid cell, and sort the obtained susceptibility probability values ​​in ascending order.

[0040] To achieve the above objectives, according to a second aspect of the present invention, an electronic device is provided, characterized in that it includes a processor and a memory for storing executable instructions of the processor;

[0041] The processor is configured to execute the aforementioned landslide susceptibility assessment method.

[0042] To achieve the above objectives, according to a third aspect of the present invention, a computer-readable storage medium is provided, characterized in that the computer-readable storage medium stores a plurality of classification programs, which are used to be called by a processor and execute the above-described landslide susceptibility assessment method.

[0043] The beneficial effects of the above-mentioned further solutions are as follows: By plotting the receiver operating characteristic curve, the present invention obtains the area under the curve value, and uses the area under the curve value to improve the prediction performance and accuracy of the machine learning model. Attached Figure Description

[0044] Figure 1 This is a flowchart of the method of the present invention.

[0045] Figure 2 This is a schematic diagram of the ROC curve in this embodiment.

[0046] Figure 3 This is a schematic diagram illustrating the performance metrics of each model in this embodiment.

[0047] Figure 4 This is the topology diagram for landslide hazard susceptibility assessment using the adaptive enhancement-categorical feature enhancement ensemble model in this embodiment.

[0048] Figure 5 This is a landslide susceptibility prediction map in this embodiment. Detailed Implementation

[0049] The specific embodiments of the present invention are described below to enable those skilled in the art to understand the present invention. However, it should be understood that the present invention is not limited to the scope of the specific embodiments. For those skilled in the art, various changes are obvious as long as they are within the spirit and scope of the present invention as defined and determined by the appended claims. All inventions utilizing the concept of the present invention are protected.

[0050] Before describing this embodiment, the following terms will be explained:

[0051] CatBoost algorithm: a categorical feature enhancement algorithm;

[0052] AdaBoost model: Adaptive boosting model;

[0053] CatBoostClassifier: A categorical feature boosting classifier;

[0054] AdaBoostClassifier: Adaptive boosting classifier;

[0055] ROC curve: Receiver operating characteristic curve;

[0056] ArcGIS: Geospatial platform;

[0057] AUC: Area under the curve;

[0058] NDVI: Normalized Difference Vegetation Index;

[0059] DEM: Digital Elevation Model;

[0060] TWI: Terrain Humidity Index.

[0061] Example 1

[0062] like Figure 1 As shown, this invention provides a landslide susceptibility assessment method based on categorical feature enhancement, the implementation method of which is as follows:

[0063] S1. Preprocess the various types of data obtained from the landslide disaster study area, and export the original values ​​of all initial evaluation factors corresponding to all landslide points in the landslide disaster study area to obtain the original dataset. The specific steps are as follows:

[0064] S101. Obtain various types of data for the landslide disaster research area;

[0065] S102. According to the preset sample ratio of landslide points and non-landslide points, randomly generate non-landslide points outside the preset diameter of landslide points;

[0066] S103. Unify the spatial resolution and coordinate system of multiple types of data, and transform the unified coordinate system data into layers of the same raster size through resampling to obtain the transformed raster data. The specific steps are as follows:

[0067] S1031. Unify the spatial resolution and coordinate system of multiple types of data;

[0068] S1032. Resample the multi-class data after unifying the coordinate system, select the elevation raster as the reference image, set the input raster image to the raster image to be resampled, set the output raster dataset to the newly created raster, set the output pixels to the same elevation raster as the layer, and set the resampling to the nearest, to obtain the converted raster data.

[0069] S104. Based on the converted raster data, use the software to extract the point tool, and according to the landslide point data, extract the original data table of all initial evaluation factors, export the original values ​​of all initial evaluation factors, and obtain the original dataset.

[0070] In this embodiment, multiple types of data are obtained for the landslide disaster research area in Yuanling County, Hunan Province;

[0071] Multiple types of data from the landslide disaster research area were preprocessed. Non-landslide points were randomly generated 1 kilometer away from the landslide points according to a preset sample ratio of 1:1 between landslide points and non-landslide points.

[0072] Before processing the various types of data, it is necessary to convert the coordinate system of the collected data to WGS_1984_UTM_Zone_46N (1984 World Geodetic Coordinate System—Transverse Mercator—Northern Hemisphere Zone 46). Therefore, the various types of data are imported into ArcGIS, and the coordinate system is unified based on Data Management Tool—Projections and Transformations—Define Projection.

[0073] Import the data into ArcGIS using a unified coordinate system. Then, resample the various types of data after unifying the coordinate system using Data Management Tools—Raster—Raster Processing—Resample. Select the elevation raster as the reference image, set the Input Raster to the raster image to be resampled, set the Output Raster Dataset to "newraster1" (the resampled raster image), set the Output Cell Size to "Same as layer (elevation raster)," and set the Resample to "NEAREST." Click OK to obtain the converted raster data.

[0074] Based on the converted raster data and using the multi-value extraction to point tool in ArcGIS 10.5 software, the original data table of all initial evaluation factors is extracted according to the landslide point data, and the original values ​​of the initial evaluation factors are exported.

[0075] S2. Establish a categorical feature enhancement model using a categorical feature enhancement algorithm, and train the model using the original dataset. Use the area under the curve as the training metric, and use a grid search algorithm to search for the optimal parameters of the categorical feature enhancement model to obtain the categorical feature enhancement model with optimal parameters. The specific steps are as follows:

[0076] S201. Establish a categorical feature enhancement model by combining categorical feature enhancement algorithms, and train the categorical feature enhancement model using the original dataset. Perform a grid search using the grid search class in the scikit-learn library. The specific steps are as follows:

[0077] S2011. Create a base learner containing instances of the categorical feature enhancement model and define a parameter grid containing the maximum number of iterations and candidate learning rate values.

[0078] S2012. Create a grid search instance, passing in the categorical feature boosting classifier, parameter grid, number of folds for cross-validation, scoring criteria, and parallel processing parameters, and use the grid search class to perform a grid search on the passed parameters.

[0079] S202. Using the fit method, the area under the curve is used as the training metric to search for the optimal parameters of the categorical feature enhancement model, and the best parameter combination is obtained. Based on the best parameter combination, the categorical feature enhancement model with the best parameters is obtained.

[0080] In this embodiment, a CatBoost model is constructed using the CatBoost algorithm. The dataset is imported into the CatBoost model for training. AUC is used as the training metric, and a grid search algorithm is used to search for the optimal parameters of the CatBoost model to obtain a CatBoost model with optimal parameters, as detailed below:

[0081] Use the GridSearchCV class from the scikit-learn library to perform a grid search to optimize the parameters of the CatBoost model. The code is: from sklearn.model_selection import GridSearchCV;

[0082] The optimal parameters for the CatBoost model were searched using a grid search algorithm, with AUC as the evaluation metric. The grid search was performed on the parameters max_iter (maximum number of iterations) and learning_rate (learning rate) to obtain the optimal parameter combination. The parameter ranges for max_iter and learning_rate were (50, 600) and (0.01, 1), respectively. The optimal parameters obtained after grid search optimization were 100 for max_iter and 0.1 for learning_rate, as detailed below:

[0083] Create a CatBoostClassifier instance as the base learner, and create a CatBoost classifier instance; Code: CatBoost = CatBoostClassifier(random_state = 42);

[0084] Define a parameter grid param_grid containing a series of candidate values ​​for max_iter and learning_rate; code: param_grid={'max_iter':(50,600),'learning_rate':(0.01,1)};

[0085] Create a GridSearchCV instance, passing in a CatBoost classifier, a parameter grid, the number of folds for cross-validation (cv=5), a scoring criterion (scoring='auc'), and parameters for parallel processing (n_jobs=-1); Code: grid_search=GridSearchCV(param_grid=param_grid,cv=5,scoring='auc',n_jobs=-1);

[0086] The `fit` method is used to perform a grid search on the training set, and the optimal parameter combination is found by evaluating the AUC metric; Code: `grid_search.fit(X_train,y_train)`;

[0087] The code outputs the best parameters found and the corresponding best algorithm score; outputs the best parameters: print("Best parameters found:",grid_search.best_params_); outputs the score of the best model: print("Best model score:",grid_search.best_score_);

[0088] A CatBoost model with optimal parameters was created using the best parameter combination and predicted on the test set. The AUC was calculated. Code: y_pred = best_model.predict(X_test); The accuracy of the CatBoost model with optimal parameters on the test set was calculated. Code: test_auc = (y_pred == y_test).mean(), print("Testset auc of best model:", test_auc).

[0089] S3. Using the adaptive enhancement model and combining it with the categorical feature enhancement model with optimal parameters, we obtain the adaptive enhancement-categorical feature enhancement ensemble model, as follows:

[0090] By using the adaptive augmentation classifier in the scikit-learn library to call the adaptive augmentation model, the adaptive augmentation model is used as the meta-learner, and the categorical feature enhancement model with the optimal parameters is used as the base learner. The meta-learner and the base learner are then integrated through adaptive augmentation to obtain the adaptive augmentation-categorical feature enhancement ensemble model.

[0091] In this embodiment, the adaptive boost classifier in the scikit-learn library is used to call the adaptive boost model to perform adaptive boost ensemble on the CatBoost model with optimal parameters. The code is from sklearn.ensembleimport AdaBoostClassifier; the expression is AdaBoostClassifier(CatBoostClassifier(iterations=100,learning_rate=0.1)).

[0092] S4. Process the original dataset to obtain an optimized dataset, and use the optimized dataset to train and simulate the adaptive augmentation-categorical feature enhancement ensemble model to obtain the trained adaptive augmentation-categorical feature enhancement ensemble model. Use evaluation metrics to test the accuracy of the trained adaptive augmentation-categorical feature enhancement ensemble model. The specific steps are as follows:

[0093] S401. Process the original dataset by setting the evaluation factor values ​​(excluding ID and attribute data Y) to X and setting the attribute data Y to the training target Y to obtain the optimized dataset.

[0094] S402. Randomly shuffle the optimized dataset according to the preset set ratio and divide it into a training set and a test set. Substitute the training set into the adaptive enhancement-categorical feature enhancement ensemble model for training simulation to obtain the trained adaptive enhancement-categorical feature enhancement ensemble model.

[0095] S403. Plot the receiver operating characteristic curve using the test set, obtain the area under the curve value, and import the area under the curve function code into the adaptive augmentation-categorical feature enhancement ensemble model. Use the area under the curve value to test the accuracy of the trained adaptive augmentation-categorical feature enhancement ensemble model.

[0096] In this embodiment, the dataset is processed using Python software. The evaluation factor values ​​other than ID and attribute data Y are set to X, while the attribute data Y is the training target Y (positive samples are 1, negative samples are 0), resulting in an optimized dataset. The optimized dataset is randomly shuffled according to a preset set ratio of 7:3 and divided into a training set and a test set. The training set is then substituted into the AdaBoost-CatBoost ensemble model for training simulation.

[0097] The model accuracy was verified using test set data, and the ROC curve was plotted to obtain the following results: Figure 2 The area under the curve (AUC) values ​​shown reveal that the AdaBoost ensemble model, with CatBoost as the base learner, significantly increases the AUC value, achieving 0.8970, which is superior to CatBoost's 0.7930 and AdaBoost's 0.6768. Performance tests were also conducted on four metrics: AUC, ACC, Precision, and F1, yielding the following results: Figure 3 The diagram showing the four performance metrics ultimately reveals that the Adaboost-CatBoost ensemble model performs best in each metric.

[0098] S5. Based on the trained adaptive augmentation-categorical feature enhancement ensemble model, generate multiple decision trees using the categorical feature enhancement classifier and predict grid points within the landslide hazard study area. Then, use the adaptive augmentation classifier to summarize the predictions and output the susceptibility probability value for each grid cell. The specific steps are as follows:

[0099] S501. Based on the trained adaptive enhancement-categorical feature enhancement ensemble model, the input layer receives the evaluation factors of the study area, and the adaptive enhancement classifier combines the base learners of multiple categorical feature enhancement classifiers.

[0100] S502. Based on the optimal parameters, use categorical features to boost the classifier to generate multiple decision trees, and predict the grid points in the landslide disaster study area based on the evaluation factors of the study area.

[0101] S503. Use an adaptive augmentation classifier to summarize the prediction results of the decision tree, output the susceptibility probability value corresponding to each grid cell, and sort the obtained susceptibility probability values ​​in ascending order.

[0102] In this embodiment, as Figure 4 As shown, the input layer: the adaptive enhancement-categorical feature enhancement ensemble model receives 16 evaluation factor information as input to the input layer; each evaluation factor is used to assess landslide susceptibility, including data on geology, topography, climate and vegetation, etc.

[0103] Adaptive Augmentation Classifier: The evaluation factors received from the input layer are fed into the adaptive augmentation classifier. The adaptive augmentation classifier is an ensemble learning method that improves the overall model performance by combining the predictions of multiple base learners that use categorical feature boosting classifiers.

[0104] Categorical Feature Boosting Classifier: The Adaptive Boosting Classifier uses a categorical feature boosting classifier as the base learner. The categorical feature boosting classifier is a gradient boosting decision tree-based algorithm configured in the Adaptive Boosting-Categorical Feature Boosting ensemble model with the optimal parameter combination of 100 iterations and a learning rate of 0.1.

[0105] Categorical feature boosting classifiers generate multiple decision trees, each of which is a base learner. They are trained independently on the input data and output prediction results.

[0106] Decision tree generation: Multiple decision trees are generated by the categorical feature boosting classifier, including the first decision tree, the second decision tree, the third decision tree, and so on up to the Nth decision tree. Each decision tree is independently trained based on the input evaluation factor information, and they together constitute the base learner set of the model.

[0107] Output layer: The predictions from all decision trees are aggregated by an adaptive booster classifier, and its weights are dynamically adjusted based on the performance of each base learner to produce the final prediction result.

[0108] In this embodiment, the adaptive enhancement-categorical feature enhancement ensemble model outputs two main results: AUC performance: a metric that measures the model's prediction accuracy; the higher the AUC value, the stronger the model's discriminative ability.

[0109] Landslide Susceptibility Mapping: The final output of the adaptive augmentation-categorical feature enhancement ensemble model generates a landslide susceptibility map based on the model's predictions. This map helps researchers and policymakers identify areas more prone to landslides and implement appropriate prevention and response measures.

[0110] This embodiment demonstrates the complete path from input data to final output, and how the various components in the model interact to improve the accuracy of landslide susceptibility assessment. This ensemble learning approach fully leverages the predictive power of multiple base learners, combining their predictions to enhance the overall model's performance and robustness.

[0111] S6. Use the natural breakpoint method to process the susceptibility probability value, obtain the landslide susceptibility evaluation result, and output the landslide susceptibility prediction map.

[0112] In this embodiment, the natural breakpoint method is used to divide the susceptibility probability value into 5 levels: low, low, medium, high, and high. The low, low, medium, high, and high susceptibility levels are automatically determined based on the actual situation, resulting in the following: Figure 5 The landslide susceptibility prediction map shown is presented. Overall, the landslide susceptibility assessment performance of the model presented in this paper is excellent. Firstly, the areas of extremely low, low, medium, high, and extremely high susceptibility zones account for 41.0%, 15.8%, 11.7%, 11.8%, and 20.1% of the total study area, respectively. Among the landslide geological hazards that have already occurred, the landslide frequency ratio reaches 6.74. Secondly, the model's AUC value is 0.897, indicating that the model's predictive performance is excellent.

[0113] Example 2

[0114] In this embodiment, the present invention provides a landslide susceptibility assessment method based on categorical feature enhancement, which has been successfully applied to landslide disaster prediction in Yuanling County, Hunan Province. With its robustness and self-learning ability, it provides a new and efficient model for landslide susceptibility assessment in Yuanling County. By optimizing the parameters of the CatBoost search algorithm and further adaptive enhancement integration, the accuracy of the geological disaster susceptibility assessment results has been improved.

[0115] In this embodiment, Yuanling County was selected as the study area. Data from the local geological bureau, the Resource and Environmental Science and Data Center of the Chinese Academy of Sciences, and Landslides_SL were integrated to obtain 289 landslides. These were randomly divided into training and validation data groups with a preset ratio of 7:3. Based on professional knowledge, 16 environmental factors were studied: elevation, slope, aspect, surface dissection coefficient, profile curvature, elevation coefficient of variation, ground roughness, curvature, TWI, river distance, road distance, fault distance, normalized difference vegetation index (NDVI), land use type, lithology, and rainfall. The digital elevation model (DEM) data came from the geospatial data cloud. The environmental factors directly obtained from this DEM were elevation, slope, aspect, surface dissection coefficient, profile curvature, elevation coefficient of variation, ground roughness, curvature, and TWI. Specifically, lithology and fault distance were obtained from the China Geological Survey's geological cloud, road and hydrological data, rainfall, and NDVI were obtained from the geospatial data cloud, and land use type data were obtained from data published by Wuhan University.

Claims

1. A landslide susceptibility assessment method based on categorical feature enhancement, characterized in that, Includes the following steps: S1. Preprocess the various types of data obtained from the landslide disaster study area, and export the original values ​​of all initial evaluation factors corresponding to all landslide points in the landslide disaster study area to obtain the original dataset. S2. Establish a categorical feature enhancement model using a categorical feature enhancement algorithm, and train the categorical feature enhancement model using the original dataset. Use the area under the curve as the training metric, and use a grid search algorithm to search for the optimal parameters of the categorical feature enhancement model to obtain a categorical feature enhancement model with optimal parameters. S3. Using the adaptive enhancement model and combining it with the categorical feature enhancement model with optimal parameters, we obtain the adaptive enhancement-categorical feature enhancement ensemble model. S4. Process the original dataset to obtain an optimized dataset, and use the optimized dataset to train and simulate the adaptive augmentation-categorical feature enhancement ensemble model to obtain the trained adaptive augmentation-categorical feature enhancement ensemble model. Use evaluation metrics to test the accuracy of the trained adaptive augmentation-categorical feature enhancement ensemble model. S5. Based on the trained adaptive enhancement-categorical feature enhancement ensemble model, use the categorical feature enhancement classifier to generate multiple decision trees and predict the grid points in the landslide disaster research area. Then, use the adaptive enhancement classifier to summarize the predictions and output the susceptibility probability value corresponding to each grid cell. S6. Use the natural breakpoint method to process the susceptibility probability value, obtain the landslide susceptibility evaluation result, and output the landslide susceptibility prediction map.

2. The landslide susceptibility assessment method based on categorical feature enhancement according to claim 1, characterized in that, S1 includes the following steps: S101. Obtain various types of data for the landslide disaster research area; S102. According to the preset sample ratio of landslide points and non-landslide points, randomly generate non-landslide points outside the preset diameter of landslide points; S103. Unify the spatial resolution and coordinate system of multiple types of data, and transform the multiple types of data after unifying the coordinate system into layers of the same raster size through resampling to obtain the transformed raster data. S104. Based on the converted raster data, use the software to extract the point tool, and according to the landslide point data, extract the original data table of all initial evaluation factors, export the original values ​​of all initial evaluation factors, and obtain the original dataset.

3. The landslide susceptibility assessment method based on categorical feature enhancement according to claim 2, characterized in that, S103 includes the following steps: S1031. Unify the spatial resolution and coordinate system of multiple types of data; S1032. Resample the multi-class data after unifying the coordinate system, select the elevation raster as the reference image, set the input raster image to the raster image to be resampled, set the output raster dataset to the newly created raster, set the output pixels to the same elevation raster as the layer, and set the resampling to the nearest, to obtain the converted raster data.

4. The landslide susceptibility assessment method based on categorical feature enhancement according to claim 1, characterized in that, S2 includes the following steps: S201. Establish a categorical feature enhancement model by combining the categorical feature enhancement algorithm, and train the categorical feature enhancement model using the original dataset. Perform grid search using the grid search class in the scikit-learn library. S202. Using the fit method, the area under the curve is used as the training metric to search for the optimal parameters of the categorical feature enhancement model, and the best parameter combination is obtained. Based on the best parameter combination, the categorical feature enhancement model with the best parameters is obtained.

5. The landslide susceptibility assessment method based on categorical feature enhancement according to claim 4, characterized in that, S201 includes the following steps: S2011. Create a base learner containing instances of the categorical feature enhancement model and define a parameter grid containing the maximum number of iterations and candidate learning rate values. S2012. Create a grid search instance, passing in the categorical feature boosting classifier, parameter grid, number of folds for cross-validation, scoring criteria, and parameters for parallel processing, and use the grid search class to perform a grid search on the passed parameters.

6. The landslide susceptibility assessment method based on categorical feature enhancement according to claim 1, characterized in that, The specific steps of S3 are as follows: By using the adaptive augmentation classifier in the scikit-learn library to call the adaptive augmentation model, the adaptive augmentation model is used as the meta-learner, and the categorical feature enhancement model with the optimal parameters is used as the base learner. The meta-learner and the base learner are then integrated through adaptive augmentation to obtain the adaptive augmentation-categorical feature enhancement ensemble model.

7. The landslide susceptibility assessment method based on categorical feature enhancement according to claim 1, characterized in that, The specific steps of S4 are as follows: S401. Process the original dataset by setting the evaluation factor values ​​(excluding ID and attribute data Y) to X and setting the attribute data Y to the training target Y to obtain the optimized dataset. S402. Randomly shuffle the optimized dataset according to the preset set ratio and divide it into a training set and a test set. Substitute the training set into the adaptive enhancement-categorical feature enhancement ensemble model for training simulation to obtain the trained adaptive enhancement-categorical feature enhancement ensemble model. S403. Plot the receiver operating characteristic curve using the test set, obtain the area under the curve value, and import the area under the curve function code into the adaptive augmentation-categorical feature enhancement ensemble model. Use the area under the curve value to test the accuracy of the trained adaptive augmentation-categorical feature enhancement ensemble model.

8. The landslide susceptibility assessment method based on categorical feature enhancement according to claim 1, characterized in that, The specific steps of S5 are as follows: S501. Based on the trained adaptive enhancement-categorical feature enhancement ensemble model, the input layer receives the evaluation factors of the study area, and the adaptive enhancement classifier combines the base learners of multiple categorical feature enhancement classifiers. S502. Based on the optimal parameters, use categorical features to boost the classifier to generate multiple decision trees, and predict the grid points in the landslide disaster study area based on the evaluation factors of the study area. S503. Use an adaptive augmentation classifier to summarize the prediction results of the decision tree, output the susceptibility probability value corresponding to each grid cell, and sort the obtained susceptibility probability values ​​in ascending order.

9. An electronic device, characterized in that, Includes a processor and memory for storing the processor's executable instructions; The processor is configured to implement the landslide susceptibility assessment method as described in any one of claims 1-8 by executing executable instructions.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a plurality of classification programs, which are used by a processor to call and execute the landslide susceptibility assessment method as described in any one of claims 1-8.

Images