Landslide susceptibility evaluation method and device, electronic equipment and medium

Abstract

The invention relates to a landslide susceptibility evaluation method and device, electronic equipment and a medium, and belongs to the technical field of geological disaster prediction and prevention, and the method comprises the steps: obtaining preprocessed digital elevation model data and multi-source data of a target area; dividing the preprocessed digital elevation model data based on preset geographic software to obtain evaluation unit data, the evaluation unit data including slope unit data and grid unit data; training a preset deep integration model based on the evaluation unit data, and determining the evaluation unit data with high training precision as target evaluation unit data; determining a landslide influence factor based on the target evaluation unit data and the multi-source data; inputting the landslide influence factor into a preset depth integration model to obtain a preliminary evaluation result of the landslide susceptibility of the target area; and obtaining a landslide susceptibility evaluation result of the target area based on the preliminary evaluation result. According to the invention, landslide disaster risk zoning and prevention and control schemes can be accurately recommended.

Description

Technical Field

[0001] This invention relates to the field of geological disaster prediction and prevention technology, and in particular to a landslide susceptibility assessment method, device, electronic equipment and medium. Background Technology

[0002] Landslides, as a common natural geological hazard, are characterized by their suddenness and destructive power, seriously threatening human life and property safety and socio-economic development. Landslide susceptibility assessment, as a key component of disaster risk management, predicts the probability of landslide occurrence by comprehensively considering multiple factors such as geology, topography, and meteorology, providing a scientific basis for disaster prevention planning and land use.

[0003] Traditional landslide susceptibility assessment methods mainly include heuristic models, statistical models, and machine learning models. However, these methods still have shortcomings in areas such as evaluation unit selection and spatial dependency modeling, which limit the accuracy and practicality of the assessment results. In recent years, the development of deep learning and ensemble learning technologies has provided new approaches to solving these problems. Convolutional neural networks (CNNs) can effectively extract local features from spatial data such as terrain and environment; ensemble learning (such as Stacking) can combine the advantages of multiple base models to improve prediction accuracy and stability. However, how to combine these advanced models with prior knowledge and engineering specifications in the field of geological hazards to construct a hybrid intelligent recommendation system that can fully utilize the potential of data while ensuring the interpretability of results and conforming to geoscientific mechanisms remains a gap and challenge in current research.

[0004] Therefore, there is an urgent need for a landslide disaster risk prevention and control scheme recommendation method that can deeply integrate multi-source data, couple multiple advanced models, and fully consider spatial dependencies, so as to provide more efficient, accurate, and scientific recommendations for landslide disaster risk zoning and prevention and control schemes. Summary of the Invention

[0005] In view of this, it is necessary to provide a landslide susceptibility assessment method, device, electronic equipment and medium to solve the problem of low accuracy in regional geological disaster prevention and control decisions.

[0006] To address the aforementioned problems, in a first aspect, the present invention provides a landslide susceptibility assessment method, comprising: Acquire preprocessed digital elevation model data and multi-source data for the target area. The multi-source data includes geological map data, rainfall data, road and river vector data, and remote sensing image data. The preprocessed digital elevation model data is divided based on the preset geographic software to obtain evaluation unit data, which includes slope unit data and raster unit data. The preset deep ensemble models are trained based on the evaluation unit data. The evaluation unit data with high training accuracy is determined as the optimal evaluation unit data. The optimal evaluation unit data is then integrated to obtain the target evaluation unit data. The preset deep ensemble model package includes at least two models from CNN model, TabTransformer, Random Forest model, and Decision Tree model. Determine landslide influencing factors based on target evaluation unit data and multi-source data; By inputting landslide influencing factors into a pre-defined deep ensemble model, preliminary evaluation results of landslide susceptibility in the target area are obtained. The landslide susceptibility assessment results for the target area were obtained based on the preliminary evaluation results.

[0007] In one possible implementation, the preset geographic software includes: One or more of GRASS GIS, QGIS, GVSIG, and SAGA GIS.

[0008] In one possible implementation, the preprocessed digital elevation model data is divided using preset geographic software to obtain evaluation unit data, including: The r.slopeunits tool in GRASS GIS was used to divide the preprocessed digital elevation model data into evaluation unit data. In one possible implementation, training a preset deep ensemble model based on evaluation unit data, and determining the evaluation unit data with high training accuracy as the target evaluation unit data, includes: When the accuracy of training the preset deep ensemble model with the slope unit data is higher than the accuracy of training the preset deep ensemble model with the grid unit data, the target evaluation unit data is determined to be slope unit data. When the accuracy of training the preset deep ensemble model with the grid cell data is higher than the accuracy of training the preset deep ensemble model with the slope cell data, the target evaluation cell data is determined to be grid cell data.

[0009] In one possible implementation, the landslide impact factors include qualitative impact factors and qualitative impact factors. The step of inputting the landslide impact factors into a pre-defined deep ensemble model to obtain preliminary evaluation results of the landslide susceptibility of the target area includes: The qualitative impact factors are labeled and coded to obtain numerical qualitative impact factor data; The quantitative impact factors are normalized to obtain standardized quantitative impact factor data. Numerical qualitative impact factor data and numerical qualitative impact factor data are input into the base learner of the preset deep ensemble model to obtain preliminary prediction data; The preliminary prediction data are combined to obtain a combined feature vector; The combined feature vectors are input into the meta-learner of the pre-defined deep ensemble model to obtain preliminary evaluation results of the landslide susceptibility of the target area.

[0010] In one possible implementation, obtaining the landslide susceptibility assessment result for the target area based on the preliminary assessment results includes: When the number of the preset deep integrated models is 1, the preliminary evaluation result is the landslide susceptibility evaluation result of the target area; When the number of preset deep integration models is not 1, the landslide susceptibility evaluation result of the target area is obtained by weighted summation based on the preliminary evaluation results.

[0011] In one possible implementation, the landslide impact factors include: Elevation, slope, aspect, plane curvature, profile curvature, topographic humidity index, lithology, distance from fault, average annual rainfall, distance from river, land use type, normalized vegetation index, and distance from road.

[0012] Secondly, the present invention also provides a landslide susceptibility assessment device, comprising: The data acquisition module is used to acquire preprocessed digital elevation model data and multi-source data of the target area. The multi-source data includes geological map data, rainfall data, road and river vector data, and remote sensing image data. The evaluation unit acquisition module is used to divide the preprocessed digital elevation model data based on preset geographic software to obtain evaluation unit data, which includes slope unit data and raster unit data. The target evaluation unit determination module is used to train the preset deep ensemble model based on the evaluation unit data, determine the evaluation unit data with high training accuracy as the optimal evaluation unit data, and integrate the optimal evaluation unit data to obtain the target evaluation unit data. The preset deep ensemble model package includes at least two models from CNN model, TabTransformer, random forest model, and decision tree model. The landslide impact factor acquisition module is used to determine landslide impact factors based on target evaluation unit data and multi-source data; The preliminary evaluation module is used to input landslide influencing factors into a preset deep ensemble model to obtain preliminary evaluation results of the landslide susceptibility of the target area; The susceptibility assessment module is used to obtain landslide susceptibility assessment results for the target area based on the preliminary assessment results.

[0013] Thirdly, the present invention also provides an electronic device, including a memory and a processor, wherein, The memory is used to store programs; The processor, coupled to the memory, is used to execute the program stored in the memory to implement the steps in the landslide susceptibility assessment method described in any of the above implementations.

[0014] Fourthly, the present invention also provides a computer-readable storage medium for storing a computer-readable program or instructions, which, when executed by a processor, can implement the steps in the landslide susceptibility assessment method described in any of the above implementations.

[0015] The beneficial effects of this invention are as follows: The landslide susceptibility assessment method provided by this invention first acquires preprocessed digital elevation model (DEM) data and multi-source data of the target area. The multi-source data includes geological map data, rainfall data, road and river vector data, and remote sensing image data. Then, based on preset geographic software, the preprocessed DEM data is divided into evaluation unit data, which includes slope unit data and raster unit data. Furthermore, a preset deep ensemble model is trained based on the evaluation unit data. The deep ensemble model includes at least one combination form, and the training uses highly accurate evaluation unit data... The optimal evaluation unit data is determined and integrated to obtain target evaluation unit data. Further, landslide influencing factors are determined based on the target evaluation unit data and multi-source data, and then fused to obtain landslide influencing factors. This accurately identifies high-risk landslide areas. The landslide influencing factors are then input into a pre-set deep ensemble model to obtain preliminary evaluation results of the landslide susceptibility of the target area. Through deep ensemble learning, by comprehensively utilizing multi-source spatial data, the advantages of unit partitioning, and the complementary characteristics of the model, a high-precision and high-stability evaluation of landslide susceptibility is achieved. Finally, based on the preliminary evaluation results, the final landslide susceptibility evaluation result for the target area is obtained. This invention determines landslide influencing factors by fusing target evaluation unit data and multi-source data, thereby accurately identifying high-risk landslide areas. Furthermore, through deep ensemble learning, by comprehensively utilizing multi-source spatial data, the advantages of unit partitioning, and the complementary characteristics of the model, a high-precision and high-stability evaluation of landslide susceptibility is achieved. Attached Figure Description

[0016] Figure 1 A flowchart illustrating an embodiment of the landslide susceptibility assessment method provided by the present invention; Figure 2 A detailed flowchart of an embodiment of the landslide susceptibility assessment method provided by the present invention; Figure 3 For the present invention Figure 1 A schematic diagram of an embodiment of S105; Figure 4This is a schematic diagram of a deep integration model combination for an embodiment of a landslide susceptibility assessment method provided by the present invention; Figure 5 An embodiment of the landslide susceptibility assessment method provided by the present invention is a landslide hazard susceptibility assessment map of a CNN-GBDT model grid cell; Figure 6 An embodiment of the landslide susceptibility assessment method provided by the present invention is a landslide susceptibility assessment map of a slope unit using a CNN-GBDT model; Figure 7 This is a schematic flowchart of an embodiment of a landslide susceptibility assessment device provided by the present invention; Figure 8 A schematic diagram of an embodiment of the electronic device provided by the present invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0018] In the description of the embodiments of the present invention, unless otherwise stated, "multiple" means two or more. "And / or" describes the relationship between related objects, indicating that there can be three relationships. For example, A and / or B can represent three situations: A exists alone, A and B exist simultaneously, and B exists alone.

[0019] The terms "first," "second," etc., used in the embodiments of this invention are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a technical feature defined with "first" or "second" may explicitly or implicitly include at least one of that feature.

[0020] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0021] Example 1: This invention provides a method, apparatus, electronic device, and medium for evaluating landslide susceptibility, which are described below.

[0022] Figure 1 This is a schematic flowchart of an embodiment of the landslide susceptibility assessment method provided by the present invention, as shown below. Figure 1 As shown, the landslide susceptibility assessment methods include: S101. Acquire preprocessed digital elevation model data and multi-source data for the target area. The multi-source data includes geological map data, rainfall data, road and river vector data, and remote sensing image data. Digital elevation models (DEMs) are spatial data models that describe the morphological features of the Earth's surface. They consist of a matrix of elevation values ​​from regular grid points on the ground, forming a raster-structured dataset.

[0023] S102. Based on the preset geographic software, the preprocessed digital elevation model data is divided to obtain evaluation unit data, which includes slope unit data and raster unit data. S103. Train the preset deep ensemble model based on the evaluation unit data, determine the evaluation unit data with high training accuracy as the optimal evaluation unit data, and integrate the optimal evaluation unit data to obtain the target evaluation unit data. The preset deep ensemble model package includes at least two models from CNN model, TabTransformer, random forest model, and decision tree model. S104. Determine landslide influencing factors based on target evaluation unit data and multi-source data; S105. Input the landslide influencing factors into the preset deep ensemble model to obtain the preliminary evaluation results of the landslide susceptibility of the target area; S106. Based on the preliminary evaluation results, the landslide susceptibility evaluation results for the target area are obtained.

[0024] Compared with existing technologies, the landslide susceptibility assessment method provided in this embodiment first acquires preprocessed digital elevation model (DEM) data and multi-source data of the target area. The multi-source data includes geological map data, rainfall data, road and river vector data, and remote sensing image data. Then, based on preset geographic software, the preprocessed DEM data is divided into evaluation unit data, which includes slope unit data and raster unit data. Further, a preset deep ensemble model is trained based on the evaluation unit data. The deep ensemble model includes at least one combination form, and the training uses highly accurate evaluation unit data. The optimal evaluation unit data is determined and integrated to obtain target evaluation unit data. Further, landslide influencing factors are determined based on the target evaluation unit data and multi-source data, and then fused to obtain landslide influencing factors. This accurately identifies high-risk landslide areas. The landslide influencing factors are then input into a pre-set deep ensemble model to obtain preliminary evaluation results of the landslide susceptibility of the target area. Through deep ensemble learning, by comprehensively utilizing multi-source spatial data, the advantages of unit partitioning, and the complementary characteristics of the model, a high-precision and high-stability evaluation of landslide susceptibility is achieved. Finally, based on the preliminary evaluation results, the final landslide susceptibility evaluation result for the target area is obtained. This invention determines landslide influencing factors by fusing target evaluation unit data and multi-source data, thereby accurately identifying high-risk landslide areas. Furthermore, through deep ensemble learning, by comprehensively utilizing multi-source spatial data, the advantages of unit partitioning, and the complementary characteristics of the model, a high-precision and high-stability evaluation of landslide susceptibility is achieved. In specific embodiments of the present invention, such as Figure 2 The following is a detailed flowchart of the method in this embodiment.

[0025] In a specific embodiment of the present invention, step S101 involves the acquisition and preprocessing of digital elevation model data and multi-source data. Specifically, taking Baoji City as an example, the digital elevation model data is 30-meter resolution ASTER GDEM digital elevation model data. The multi-source data includes: 1:200,000 scale geological map data, 1-kilometer resolution annual average rainfall data, Landsat 8 remote sensing image data, OpenStreetMap road and river vector data, and historical landslide cataloging data obtained through multi-temporal high-resolution remote sensing image interpretation. Strict quality control, format standardization, and spatial registration are performed on the digital elevation model data and multi-source data to ensure data consistency and accuracy.

[0026] Specifically, the historical landslide cataloging data comes from the visual interpretation of multi-temporal high-resolution remote sensing images from March 2009 to March 2022, including Quickbird, IKONOS, and Pleiades imagery, with a maximum resolution of 0.5 meters. A total of 3,422 large landslides with an area greater than 5,000 square meters were identified, and a complete landslide cataloging database was established. All historical landslide cataloging data underwent ArcGIS topology checks to ensure the accuracy and reliability of the data.

[0027] In some embodiments of the present invention, in step S102, the preset geographic software includes: One or more of GRASS GIS, QGIS, GVSIG, and SAGA GIS.

[0028] In some embodiments of the present invention, in step S102, the preprocessed digital elevation model data is divided based on preset geographic software to obtain evaluation unit data, including: The r.slopeunits tool in GRASS GIS was used to divide the preprocessed digital elevation model data into evaluation unit data. In some embodiments of the present invention, the preset deep ensemble model includes a base learner and a meta learner. The base learner includes one or more of the following: random forest model, gradient boosting decision tree model, convolutional neural network model, and TabTransformer model. The meta learner includes one or more of the following: logistic regression model, linear model, and simple tree model.

[0029] In some embodiments of the present invention, the step of training a preset deep ensemble model based on evaluation unit data, and determining the evaluation unit data with high training accuracy as the target evaluation unit data, includes: When the accuracy of training the preset deep ensemble model with the slope unit data is higher than the accuracy of training the preset deep ensemble model with the grid unit data, the target evaluation unit data is determined to be slope unit data. When the accuracy of training the preset deep ensemble model with the grid cell data is higher than the accuracy of training the preset deep ensemble model with the slope cell data, the target evaluation cell data is determined to be grid cell data.

[0030] In a specific embodiment of the present invention, step S102 specifically includes: multi-scale evaluation unit division and optimization. More specifically, based on the processed digital elevation model data, the r.slopeunits tool in GRASS GIS is used for adaptive division of slope units. The optimal parameter combination is determined through an iterative optimization algorithm, setting the minimum unit area parameter areamin to 500,000 square meters and the minimum circular variance parameter cymin to 0.35. After 10 iterations of optimization, 15,364 topographically reasonable slope units are generated. Simultaneously, standard 30m × 30m raster units are generated as the basis for comparative analysis, forming a complete dual evaluation unit system.

[0031] More specifically, the r.slopeunits tool's partitioning principle is based on topographic and hydrological theories, including the following key steps: First, based on watershed and catchment line regional segmentation, watersheds and catchment lines are extracted from DEM data, initially dividing the study area into a series of "semi-watershed" units; second, the partitioning results are optimized by measuring the geomorphic homogeneity (such as slope aspect consistency) within units and the geomorphic discontinuity between units, using circular variance (CV) to quantify the uniformity of slope aspect within units, and ensuring the physical rationality of the partitioning results in terms of morphology by limiting the CV value; third, the partitioning results are refined through multiple iterations. In the initial partitioning, a larger catchment area threshold is used to generate coarser units, and as the iteration progresses, the threshold is gradually reduced, making the unit partitioning more detailed.

[0032] Multi-resolution mesh generation and spatial autocorrelation analysis are performed on the divided grid cells. Geometric quality assessment is conducted on the divided slope cells. The two cell datasets are input into a deep ensemble model for training and prediction, respectively, to obtain accuracy metrics (such as AUC and F1 scores). Automated decision-making and output are performed, comparing the model accuracy of the two cell types and conducting significance tests.

[0033] Condition 1: If the accuracy of the ramp element (SUI) is significantly higher than that of the grid element (RUI) and the geometric quality meets the standard, then the ramp element is adopted as the evaluation element.

[0034] Condition 2: If the above conditions are not met, the system will default to using a single grid cell as the evaluation unit.

[0035] Condition 3: If the difference in accuracy between the two is not statistically significant, then the choice should be made based on application requirements and preferences.

[0036] In some embodiments of the present invention, the landslide influencing factors include: Elevation, slope, aspect, plane curvature, profile curvature, topographic humidity index, lithology, distance from fault, average annual rainfall, distance from river, land use type, normalized vegetation index, and distance from road.

[0037] In a specific embodiment of this invention, the landslide influencing factor system is constructed and optimized as follows: 13 core influencing factors are systematically extracted from four dimensions: topography, geological environment, hydrology and meteorology, and human activities. These factors include elevation, slope, aspect, plane curvature, profile curvature, topographic humidity index (TWI), lithology, distance from fault, average annual rainfall, distance from river, land use type, normalized difference vegetation index (NDVI), and distance from road. Pearson correlation coefficient and information gain ratio are used for factor screening to eliminate redundant factors with high correlation but low importance, ensuring the scientific validity and representativeness of the factor system.

[0038] Specifically, frequency ratio analysis was used to analyze the correlation between different levels of various factors and landslide occurrence. The results showed that among the elevation factors, the range of 958.7–1397.6 meters had the highest frequency ratio (1.47); among the slope factors, the range of 10.84–19.58° had the highest frequency ratio (1.91); among the lithology factors, sedimentary rocks had the highest frequency ratio (1.26); and among the rainfall factors, the range of 638.5–700.5 mm had the highest frequency ratio (1.602). These results provide a scientific basis for factor selection and weight determination.

[0039] In some embodiments of the present invention, obtaining the landslide susceptibility assessment result of the target area based on the preliminary evaluation results includes: When the number of the preset deep integrated models is 1, the preliminary evaluation result is the landslide susceptibility evaluation result of the target area; When the number of preset deep integration models is not 1, the landslide susceptibility evaluation result of the target area is obtained by weighted summation based on the preliminary evaluation results.

[0040] In some embodiments of the present invention, in step S105, such as Figure 3 As shown, the landslide influencing factors include qualitative influencing factors and qualitative influencing factors. The process of inputting these landslide influencing factors into a pre-defined deep ensemble model to obtain preliminary evaluation results of the landslide susceptibility of the target area includes: S301. Perform label coding on the qualitative impact factors to obtain numerical qualitative impact factor data; S302. Normalize the quantitative impact factors to obtain standardized quantitative impact factor data; S303. Input the numerical qualitative impact factor data and the numerical qualitative impact factor data into the base learner of the preset deep ensemble model to obtain preliminary prediction data. S304. Combine the preliminary prediction data to obtain a combined feature vector; S305. Input the combined feature vector into the meta-learner of the preset deep ensemble model to obtain the preliminary evaluation results of the landslide susceptibility of the target area.

[0041] In a specific embodiment of the present invention, the topographic factor data in each slope unit is preprocessed and format converted. The multidimensional topographic factors such as elevation, slope, aspect, plane curvature, profile curvature, and topographic humidity index are reorganized according to their spatial distribution characteristics and converted into a two-dimensional feature map format. The size of each feature map matches the geometric shape of the corresponding slope unit, ensuring the complete preservation and effective expression of topographic spatial information. Local spatial features are extracted using convolutional layers, pooling layers, and fully connected layers. The feature tensors obtained after convolution and pooling operations are flattened into one-dimensional feature vectors and input into fully connected layers for the synthesis and reorganization of advanced terrain features. Through linear transformation of weight matrices and bias vectors combined with nonlinear mapping of ReLU activation function, deep abstraction and representation of complex terrain features are achieved.

[0042] In a specific embodiment of the present invention, a multi-level deep ensemble learning model system is constructed based on the screened influence factor system: First, a baseline model is established, including Random Forest (RF) and Gradient Boosting Decision Tree (GBDT); second, a deep learning model is constructed, including Convolutional Neural Network (CNN) and TabTransformer; then, a deep ensemble model is developed, employing a Stacking ensemble strategy to construct combined models such as CNN-RF, CNN-GBDT, TabTransformer-RF, and TabTransformer-GBDT, for example... Figure 4 As shown.

[0043] Specifically, the model training uses a 5-fold cross-validation method, which randomly divides the influence factor system after the dataset is selected into 5 parts, and uses 4 parts of the data for training and 1 part of the data for validation in turn, and repeats 5 times to ensure that all data participate in training and validation.

[0044] The model performance is comprehensively evaluated using multiple metrics such as accuracy, precision, recall, F1 score, and AUC.

[0045] Overall accuracy reflects the proportion of samples correctly classified by the model, and its value ranges from (0, 1). The closer it is to 1, the better the model's classification ability. The calculation formula is:

[0046] Wherein, TP represents true positives, which are the number of samples that the model predicts to be positive and are actually positive; TN represents true negatives, which are the number of samples that the model predicts to be negative and are actually negative; FP represents false positives, which are the number of samples that the model predicts to be positive but are actually negative; and FN represents false negatives, which are the number of samples that the model predicts to be negative but are actually positive.

[0047] Precision measures the proportion of actual landslides in the units predicted as landslides by the model, reflecting the reliability of the prediction results. Its value ranges from (0, 1]. High precision indicates a low false alarm rate. The calculation formula is:

[0048] Recall rate represents the proportion of positive samples predicted as landslides by the model, and is calculated using the following formula:

[0049] The F1 score combines precision and recall, thus avoiding bias caused by a single metric. In landslide susceptibility assessment, the F1 score effectively reflects the model's ability to balance reducing false positives and false negatives, especially suitable for datasets with uneven distribution of positive and negative samples. The F1 score ranges from (0, 1], with a higher value indicating a better balance between precision and recall. The calculation formula is:

[0050] AUC measures the area under the ROC curve. The closer the AUC is to 1, the closer the ROC curve is to the top left corner. This means the model can better improve the true positive rate and reduce the false positive rate at different classification thresholds, indicating good overall model performance. The calculation formula is:

[0051] Where n represents the total number of samples; i represents the i-th sample; This refers to the true label of the i-th sample; It refers to the probability that the model predicts the i-th sample as positive.

[0052] The evaluation metrics for several models based on grid cells are shown in Table 1 below: Table 1: Evaluation Metrics for Deep Integration Models Based on Grid Cells

[0053] The evaluation indices for several models based on slope elements are shown in Table 2 below: Table 2: Evaluation Indicators for Deep Integration Model Based on Slope Elements

[0054] Experimental results show that the CNN-GBDT model performs best on the raster cell dataset, achieving an AUC of 0.972 and an F1 score of 0.968, representing a 7.71% improvement in accuracy and a 13.2% improvement in recall compared to a single CNN model. On the ramp cell dataset, the CNN-GBDT model also maintains high performance, with an AUC of 0.877. Notably, the CNN-RGCN model performs exceptionally well on the ramp cell dataset, achieving an overall accuracy of 87.82% and a recall of 91.98%, representing a 5.71% improvement over the traditional random forest model.

[0055] In a specific embodiment of the present invention, step S106 specifically includes: multi-model collaborative prediction and intelligent decision-making. Specifically, a multi-model collaborative prediction mechanism is adopted, and model predictions are performed based on grid cells and slope cells respectively to generate multiple sets of risk assessment results. Through a weighted fusion algorithm, corresponding weights are assigned according to the performance of each model on the validation set to achieve intelligent fusion of prediction results.

[0056] Specifically, the natural breaks method was used to classify the final susceptibility assessment results into five risk levels: extremely low susceptibility, low susceptibility, moderate susceptibility, high susceptibility, and extremely high susceptibility. The scientific validity of the risk classification was verified by calculating the distribution frequency ratio of historical landslide points within each level area. The analysis results show that the frequency ratio of the extremely high susceptibility area is significantly higher than that of the other levels, confirming the rationality and reliability of the classification.

[0057] Specifically, Figure 5 and Figure 6 This is a landslide susceptibility assessment map based on the CNN-GBDT model. The map shows that areas with high and extremely high landslide susceptibility are mainly distributed in the central, northeastern, northwestern, and southwestern parts of Baoji City. The assessment results obtained by the ensemble model using GBDT as one of the base classifiers show that areas with extremely low and extremely high susceptibility account for a large proportion, while areas with low, medium, and high susceptibility account for a very small proportion.

[0058] Finally, risk assessment and prevention and control decision support are provided. Specifically, based on the final risk zoning results, differentiated prevention and control measures are recommended for areas with different risk levels. For extremely high-risk areas, detailed engineering geological surveys are recommended, along with the implementation of professional monitoring and early warning systems, and consideration of necessary engineering remediation measures. For high-risk areas, patrols and monitoring are recommended, human engineering activities are restricted, and necessary protective engineering is implemented. For moderate-risk areas, community-based monitoring and prevention efforts are recommended, along with public education on geological disaster prevention and mitigation. For low and extremely low-risk areas, routine management and continuous monitoring are recommended.

[0059] Example 2: Step 1: Construct a multi-source dataset. Based on raster cells and slope cells, construct landslide susceptibility assessment datasets respectively. The datasets include four types of landslide influencing factors: topography, geological conditions, hydrology and meteorology, and human activities. This step involves preprocessing and registering multi-source spatial data, including Digital Elevation Model (DEM), geological data, rainfall data, river and road data, NDVI, and land use data, to form an initial spatial database. Further, using the r.slopeunits tool in GRASS GIS, hydrological analysis and morphological segmentation methods are employed to adaptively divide the study area into slope units, creating evaluation units that highly match the terrain continuity. Simultaneously, traditional raster units are retained as comparison units. Finally, a two-unit dataset is constructed, containing 13 influencing factors: elevation, slope, aspect, plane curvature, profile curvature, TWI, lithology, distance to fault, average annual rainfall, distance to river, land use type, NDVI, and distance to road. This provides a data foundation for subsequent model training.

[0060] Step 2: Based on the Stacking ensemble strategy, construct deep ensemble models, including four combined models: CNN-RF, CNN-GBDT, TabTransformer-RF, and TabTransformer-GBDT; This step is explained in detail, using Convolutional Neural Networks (CNN) and TabTransformer as feature extraction base models, and Random Forest (RF) and Gradient Boosting Decision Tree (GBDT) as decision base models. The Stacking ensemble method is used to combine the advantages of multiple models. Specifically, it includes: The dataset was divided into training and test sets in an 8:2 ratio, and the results were fed into base learners such as CNN, TabTransformer, RF, and GBDT for training. CNN extracted local spatial features through convolutional layers, batch normalization layers, ReLU activation functions, and max pooling layers; TabTransformer processed structured tabular data through a self-attention mechanism to capture complex relationships between features; RF and GBDT performed robust predictions by ensemble multi-decision tree and gradient boosting strategies, respectively. After each base model output its predicted probability, it was used as a meta-feature input to the logistic regression (LR) meta-learner for secondary training and prediction, ultimately outputting the ensemble prediction result.

[0061] Step 3: Train a deep ensemble model based on the grid cell and slope cell datasets respectively, and output a landslide susceptibility probability map; This step involves inputting the grid cell and slope cell datasets into the trained deep ensemble model to obtain the landslide occurrence probability for each cell. The natural breakpoint method is used to classify the probability values ​​into five susceptibility levels: extremely low, low, medium, high, and extremely high, generating a visual susceptibility distribution map. The frequency ratio (FR) is used to verify the concentration of historical landslide distribution within each level, evaluating the model's accuracy and reliability.

[0062] Step 4: Compare the evaluation results of different units and different models, integrate the advantages of multiple models, and output the final susceptibility evaluation map; This step is explained in detail, comparing the evaluation results of grid cells and slope cells under different models to analyze their spatial distribution consistency, model stability, and landslide identification ability. For areas with inconsistent results, a frequency ratio-weighted fusion strategy is introduced, prioritizing the model output with high recall and good AUC value as the final evaluation result. Finally, a landslide susceptibility distribution map integrating the advantages of multiple models is generated, providing a scientific basis for regional geological disaster prevention and land planning.

[0063] Further explanation of the embodiments of the present invention: the ensemble learning part applies the dataset to a learning model that uses logistic regression as the meta-learner and CNN, TabTransformer, RF, and GBDT as base learners in a stacked ensemble. Test and training data are generated in an 8:2 ratio to obtain prediction results, and the final evaluation results are output through susceptibility mapping and frequency ratio validation. Model parameters are automatically adjusted through hyperparameter optimization and cross-validation. Attention mechanisms and graph structure optimization are introduced for models with unsatisfactory performance. Ultimately, the highest AUC of 0.9724 is achieved on the grid cell, and the CNN-RGCN model achieves a recall rate of 91.98% on the ramp cell, significantly outperforming traditional methods.

[0064] To better implement the landslide susceptibility assessment method in this embodiment of the invention, based on a landslide susceptibility assessment method, correspondingly, as follows: Figure 7 As shown, this embodiment of the invention also provides a landslide susceptibility assessment device. A landslide susceptibility assessment device 700 includes: The data acquisition module 701 is used to acquire preprocessed digital elevation model data and multi-source data of the target area. The multi-source data includes geological map data, rainfall data, road and river vector data, and remote sensing image data. The evaluation unit acquisition module 702 is used to divide the preprocessed digital elevation model data based on preset geographic software to obtain evaluation unit data, which includes slope unit data and raster unit data. The target evaluation unit determination module 703 is used to train the preset deep ensemble model based on the evaluation unit data, determine the evaluation unit data with high training accuracy as the optimal evaluation unit data, and integrate the optimal evaluation unit data to obtain the target evaluation unit data. The preset deep ensemble model package includes at least two models from CNN model, TabTransformer, random forest model, and decision tree model. The landslide impact factor acquisition module 704 is used to determine landslide impact factors based on target evaluation unit data and multi-source data. The preliminary evaluation module 705 is used to input landslide influencing factors into a preset deep ensemble model to obtain preliminary evaluation results of the landslide susceptibility of the target area. The susceptibility assessment module 706 is used to obtain the landslide susceptibility assessment results for the target area based on the preliminary assessment results.

[0065] The landslide susceptibility assessment device 700 provided in the above embodiments can realize the technical solution described in the above embodiment of the landslide susceptibility assessment method. The specific implementation principle of each module or unit can be found in the corresponding content in the above embodiment of the landslide susceptibility assessment method, and will not be repeated here.

[0066] like Figure 8 As shown, the present invention also provides an electronic device 800. The electronic device 800 includes a processor 801, a memory 802, and a display 803. Figure 8 Only some components of the electronic device 800 are shown, but it should be understood that it is not required to implement all the components shown, and more or fewer components may be implemented instead.

[0067] In some embodiments, processor 801 may be a central processing unit (CPU), microprocessor, or other data processing chip, used to run program code stored in memory 702 or process data, such as a landslide susceptibility assessment method of the present invention.

[0068] In some embodiments, processor 801 may be a single server or a group of servers. The server group may be centralized or distributed. In some embodiments, processor 801 may be local or remote. In some embodiments, processor 801 may be implemented on a cloud platform. In some embodiments, the cloud platform may include private cloud, public cloud, hybrid cloud, community cloud, distributed cloud, internal cloud, multi-cloud, etc., or any combination thereof.

[0069] In some embodiments, memory 802 may be an internal storage unit of electronic device 800, such as a hard disk or memory of electronic device 800. In other embodiments, memory 802 may also be an external storage device of electronic device 800, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc. equipped on electronic device 800.

[0070] Furthermore, the memory 802 may include both internal storage units of the electronic device 800 and external storage devices. The memory 802 is used to store application software and various types of data installed on the electronic device 800.

[0071] In some embodiments, display 803 may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, or an OLED (Organic Light-Emitting Diode) touchscreen. Display 803 is used to display information from electronic device 800 and to display a visual user interface. Components 801-803 of electronic device 800 communicate with each other via a system bus.

[0072] In one embodiment, when processor 801 executes a landslide susceptibility assessment program stored in memory 802, the following steps can be implemented: Acquire preprocessed digital elevation model data and multi-source data for the target area. The multi-source data includes geological map data, rainfall data, road and river vector data, and remote sensing image data. The preprocessed digital elevation model data is divided based on the preset geographic software to obtain evaluation unit data, which includes slope unit data and raster unit data. The preset deep ensemble models are trained based on the evaluation unit data. The evaluation unit data with high training accuracy is determined as the optimal evaluation unit data. The optimal evaluation unit data is then integrated to obtain the target evaluation unit data. The preset deep ensemble model package includes at least two models from CNN model, TabTransformer, Random Forest model, and Decision Tree model. Determine landslide influencing factors based on target evaluation unit data and multi-source data; By inputting landslide influencing factors into a pre-defined deep ensemble model, preliminary evaluation results of landslide susceptibility in the target area are obtained. The landslide susceptibility assessment results for the target area were obtained based on the preliminary evaluation results.

[0073] It should be understood that when the processor 801 executes a landslide susceptibility evaluation program in the memory 802, in addition to the functions mentioned above, it can also perform other functions, as detailed in the description of the corresponding method embodiments above.

[0074] Furthermore, the embodiments of the present invention do not specifically limit the type of electronic device 800 mentioned. Electronic device 800 can be a mobile phone, tablet computer, personal digital assistant (PDA), wearable device, laptop computer, or other portable electronic device. Exemplary embodiments of portable electronic devices include, but are not limited to, portable electronic devices running iOS, Android, Microsoft, or other operating systems. The aforementioned portable electronic device can also be other portable electronic devices, such as a laptop computer with a touch-sensitive surface (e.g., a touch panel). It should also be understood that in some other embodiments of the present invention, electronic device 800 may not be a portable electronic device, but rather a desktop computer with a touch-sensitive surface (e.g., a touch panel).

[0075] Those skilled in the art will understand that all or part of the processes of the methods described in the above embodiments can be implemented by a computer program instructing related hardware, and the program can be stored in a computer-readable storage medium. The computer-readable storage medium may be a disk, optical disk, read-only memory, or random access memory, etc.

[0076] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for evaluating landslide susceptibility, characterized in that, include: Acquire preprocessed digital elevation model data and multi-source data for the target area. The multi-source data includes geological map data, rainfall data, road and river vector data, and remote sensing image data. The preprocessed digital elevation model data is divided based on the preset geographic software to obtain evaluation unit data, which includes slope unit data and raster unit data. The preset deep ensemble models are trained based on the evaluation unit data. The evaluation unit data with high training accuracy is determined as the optimal evaluation unit data. The optimal evaluation unit data is then integrated to obtain the target evaluation unit data. The preset deep ensemble model package includes at least two models from CNN model, TabTransformer, Random Forest model, and Decision Tree model. Determine landslide influencing factors based on target evaluation unit data and multi-source data; By inputting landslide influencing factors into a pre-defined deep ensemble model, preliminary evaluation results of landslide susceptibility in the target area are obtained. The landslide susceptibility assessment results for the target area were obtained based on the preliminary evaluation results.

2. The landslide susceptibility assessment method according to claim 1, characterized in that, The preset geographic software includes: One or more of GRASS GIS, QGIS, GVSIG, and SAGA GIS.

3. The landslide susceptibility assessment method according to claim 2, characterized in that, The preprocessed digital elevation model data is divided using preset geographic software to obtain evaluation unit data, including: The r.slopeunits tool in GRASS GIS was used to divide the preprocessed digital elevation model data into evaluation unit data.

4. The landslide susceptibility assessment method according to claim 1, characterized in that, The step of training a pre-defined deep ensemble model based on evaluation unit data, and determining the evaluation unit data with high training accuracy as target evaluation unit data, includes: When the accuracy of training the preset deep ensemble model with the slope unit data is higher than the accuracy of training the preset deep ensemble model with the grid unit data, the target evaluation unit data is determined to be slope unit data. When the accuracy of training the preset deep ensemble model with the grid cell data is higher than the accuracy of training the preset deep ensemble model with the slope cell data, the target evaluation cell data is determined to be grid cell data.

5. The landslide susceptibility assessment method according to claim 1, characterized in that, Landslide influencing factors include qualitative influencing factors and qualitative influencing factors. The process of inputting these landslide influencing factors into a pre-defined deep ensemble model yields preliminary evaluation results of the landslide susceptibility of the target area, including: The qualitative impact factors are labeled and coded to obtain numerical qualitative impact factor data; The quantitative impact factors are normalized to obtain standardized quantitative impact factor data. Numerical qualitative impact factor data and numerical qualitative impact factor data are input into the base learner of the preset deep ensemble model to obtain preliminary prediction data; The preliminary prediction data are combined to obtain a combined feature vector; The combined feature vectors are input into the meta-learner of the pre-defined deep ensemble model to obtain preliminary evaluation results of the landslide susceptibility of the target area.

6. The landslide susceptibility assessment method according to claim 1, characterized in that, The landslide susceptibility assessment results for the target area obtained based on the preliminary assessment results include: When the number of the preset deep integrated models is 1, the preliminary evaluation result is the landslide susceptibility evaluation result of the target area; When the number of preset deep integration models is not 1, the landslide susceptibility evaluation result of the target area is obtained by weighted summation based on the preliminary evaluation results.

7. The landslide susceptibility assessment method according to claim 1, characterized in that, The landslide influencing factors include: Elevation, slope, aspect, plane curvature, profile curvature, topographic humidity index, lithology, distance from fault, average annual rainfall, distance from river, land use type, normalized vegetation index, and distance from road.

8. A landslide susceptibility assessment device, characterized in that, include: The data acquisition module is used to acquire preprocessed digital elevation model data and multi-source data of the target area. The multi-source data includes geological map data, rainfall data, road and river vector data, and remote sensing image data. The evaluation unit acquisition module is used to divide the preprocessed digital elevation model data based on preset geographic software to obtain evaluation unit data, which includes slope unit data and raster unit data. The target evaluation unit determination module is used to train the preset deep ensemble model based on the evaluation unit data, determine the evaluation unit data with high training accuracy as the optimal evaluation unit data, and integrate the optimal evaluation unit data to obtain the target evaluation unit data. The preset deep ensemble model package includes at least two models from CNN model, TabTransformer, random forest model, and decision tree model. The landslide impact factor acquisition module is used to determine landslide impact factors based on target evaluation unit data and multi-source data; The preliminary evaluation module is used to input landslide influencing factors into a preset deep ensemble model to obtain preliminary evaluation results of the landslide susceptibility of the target area; The susceptibility assessment module is used to obtain landslide susceptibility assessment results for the target area based on the preliminary assessment results.

9. An electronic device, characterized in that, Including memory and processor, among which, The memory is used to store programs; The processor, coupled to the memory, is used to execute the program stored in the memory to implement the steps in the landslide susceptibility assessment method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, Used to store computer-readable programs or instructions, which, when executed by a processor, can implement the steps in the landslide susceptibility assessment method according to any one of claims 1 to 7.

Images