Landslide risk evaluation method based on convolutional neural network coupling model

Abstract

The invention discloses a landslide risk evaluation method based on a convolutional neural network coupling model, and the method comprises the steps: S1, collecting natural geological environment background data of a research region, and obtaining statistical factors affecting the formation, inoculation and occurrence of a landslide; s2, calculating the stability of the slope body in the research area through a Scoops 3D physical and mechanical model, taking the calculated landslide stability coefficient as a physical model factor, and establishing a landslide risk evaluation factor set with a statistical factor; s3, determining a landslide sample and a non-landslide sample, and a training set and a test set of each sample according to the landslide risk evaluation factor set and the historical landslide distribution data of the research area; s4, inputting the training set of each sample into a convolutional neural network model to obtain a trained coupled convolutional neural network model; and S5, inputting the test set into the trained convolutional neural network model to obtain landslide risk grading evaluation. According to the invention, the risk of landslide in the research area can be evaluated more accurately.

Description

Technical Field

[0001] The invention belongs to the technical field of landslide disaster assessment, and in particular relates to a landslide hazard assessment method based on a convolutional neural network coupling model. Background Art

[0002] At present, the quantitative evaluation methods of regional landslide hazards can be roughly divided into three categories: statistical models, deterministic models and artificial intelligence models.

[0003] Statistical models mainly use various statistical methods and theoretical models of modern mathematical statistics to establish the correlation between various factors affecting landslides and the occurrence of landslides, so as to evaluate the possibility of landslides in areas with similar factor combinations. Representative models include: information volume model, evidence weight model, determination coefficient model and frequency ratio model. This type of model can comprehensively consider the impact of multiple factors on the hazard of landslides. It is relatively simple overall. Since the evaluation factors are easy to obtain, the model is relatively easy to operate and easy to promote and apply. However, it has high requirements on the representativeness of the evaluation factors and the number and quality of samples, and generally cannot consider time factors and the physical occurrence mechanism of landslides.

[0004] The deterministic model is based on the mechanical principle of slope instability, and combines the limit equilibrium theory to establish a mechanical model. The danger of landslides is evaluated by calculating the slope stability or the probability of landslide occurrence. The physical deterministic model has gradually developed from one-dimensional and two-dimensional models to three-dimensional models. The one-dimensional and two-dimensional models are represented by the SHALSTAB model, dSLAM model, SINMAP model, and TRIGRS model. However, the above models ignore the sliding direction and three-dimensional geometric characteristics of the landslide, and have problems such as over-simplification of the landslide mechanism and difficulty in reflecting the mechanical mechanism of the landslide. The existing widely used commercial mature software such as FLAC3d, ANSYS and Abacus is only suitable for the stability analysis and prediction of three-dimensional single landslides. The Scoops3D model is a three-dimensional slope stability analysis developed by the United States Geological Survey. It has simple modeling and can be organically coupled with some quantitative models or statistical models to consider complex geological conditions. It is particularly suitable for the identification and evaluation of regional landslides. The advantage of this model is that it can take into account the mechanical mechanism of landslide occurrence and three-dimensional complex terrain and geological conditions. Its disadvantage is that it is difficult to accurately obtain the physical mechanics and hydrogeological parameters of rock and soil in a large area, and it fails to fully consider the comprehensive impact of various factors on the occurrence of landslides.

[0005] The artificial intelligence model assumes that the geological conditions and environment of current or future landslides are similar to those of historical landslides, and makes predictions by analyzing the statistical relationship between historical landslides and geological conditions. Machine learning models and deep learning models have great advantages because they can use existing data and solve complex nonlinear problems. They are currently widely used in landslide hazard assessment, such as random forests, support vector machines, convolutional neural networks, and generative adversarial networks. Among these models, the convolutional neural network model, as one of the classic supervised learning models, improves the accuracy and stability of the model by constructing multiple convolutional layers and selecting the optimal influencing factors. Case studies have shown that the convolutional neural network model has strong predictive ability, is suitable for regional landslide research, and can significantly improve the accuracy and precision of regional landslide hazard assessment results. In landslide hazard assessment, the main problem of the convolutional neural network model is its "black box nature", that is, it cannot effectively consider the failure process and instability mechanism of the landslide, so it is difficult to explain the decision-making process of the model. In addition, the model relies heavily on the quality and reliability of sample data.

[0006] As mentioned above, the three types of landslide hazard assessment models have their own advantages and disadvantages. In view of the shortcomings of the existing assessment models, in order to improve the accuracy of geological hazard assessment, combined models or coupling models have become an inevitable trend in geological hazard assessment in recent years. Summary of the invention

[0007] In view of the above-mentioned deficiencies in the prior art, the present invention provides a landslide hazard assessment method based on a convolutional neural network coupling model, which solves the problem that the existing assessment model has low accuracy in landslide hazard assessment and cannot simultaneously consider the influence of various statistical factors and physical and mechanical mechanisms.

[0008] In order to achieve the above-mentioned invention object, the technical solution adopted by the present invention is: a landslide hazard assessment method based on a convolutional neural network coupling model, comprising the following steps: S1. Collect environmental background data of the study area and obtain key statistical factors that affect the formation, development and occurrence of landslides; S2. The stability of the slope in the study area is calculated by Scoops3D physical and mechanical model, and the calculated landslide stability coefficient is used as a physical model factor to establish a landslide hazard assessment factor set with statistical factors; S3. Determine the landslide samples and non-landslide samples, as well as the training set and test set of each sample based on the landslide hazard assessment factor set and the historical landslide distribution data of the study area; S4, inputting the training set of each sample into the convolutional neural network model to obtain a trained coupled convolutional neural network model; S5. Input the test set into the trained coupled convolutional neural network model to obtain the landslide hazard classification evaluation.

[0009] Furthermore: in S1, the key statistical factors include topographic factors, geological factors, meteorological and hydrological factors, and seismic factors, among which the topographic factors include slope, aspect, and relative height difference; the geological factors include engineering geological rock groups and the distance from faults; the meteorological and hydrological factors include the distance from the water system and the average maximum daily rainfall over many years; and the seismic factor is the earthquake intensity.

[0010] Further: In S2, the landslide stability coefficient is calculated F s The specific expression is:

[0011] In the formula, i Number the rows of the 3D cylinder, j is the column number for the 3D cylinder, R ij is the search radius, e ij To search for the distance from the center of the sphere to the center of mass of the three-dimensional cylinder, A ij is the sliding surface area of ​​the potential sliding body, c ij is the cohesion of the potential sliding body, φ ij is the internal friction angle of the potential sliding body, α ij is the apparent inclination angle of the potential sliding body, W ij is the gravity of the potential sliding body, u ij is the pore water pressure, k ij is the horizontal vibration load.

[0012] Further: In S4, the convolutional neural network model includes a first convolutional layer, a first pooling layer, a second convolutional layer, a second pooling layer, a third convolutional layer, a third pooling layer, a fully connected layer and an output layer connected in sequence.

[0013] Further: In S4, the method for obtaining the trained convolutional neural network coupling model is specifically: S41, inputting each factor in the training set into the convolutional neural network model respectively, and specifying the weight ratio of the physical model factor and the statistical factor by the principal component proportion method, thereby controlling the calculation process of the convolutional neural network model; S42. Use the cross entropy loss function to calculate the error between the prediction result and the true label, and use the Adam optimizer to train the model to minimize the prediction error and obtain the trained convolutional neural network coupling model.

[0014] Further: In S41, the physical model factors account for 50% of the principal components affecting the convolutional neural network model calculation, and the statistical factors account for 50% of the principal components affecting the convolutional neural network model calculation.

[0015] Further: In S42, the cross entropy loss function L The specific expression is:

[0016] In the formula, N is the total number of samples in the dataset, y i For the i The true labels of samples, For the i The predicted output of samples; The specific expression of Adam optimizer is:

[0017] In the formula, is the updated model weight parameter, is the model weight parameter before updating, η is the learning rate, λ is the L2 regularization coefficient, is the sign of the partial derivative, w is the convolution kernel.

[0018] The beneficial effects of the present invention are: (1) The present invention provides a landslide hazard assessment method based on a convolutional neural network coupling model. Based on the investigation, remote sensing interpretation and analysis of the environmental background conditions and disaster characteristics of historical landslide disasters in the study area, the Scoops3D-CNN coupling model is innovatively proposed for quantitative assessment of landslide hazards in the study area. In this coupling model, the principal component ratio is used for the first time to adjust the weight ratio of the physical model evaluation factor and the statistical evaluation factor. Compared with the method of direct classification using a convolutional neural network, this method can find the most optimized and most suitable configuration model method for actual research.

[0019] (2) The coupling model proposed in this paper can not only comprehensively consider the impact of various factors on the incubation and occurrence of landslides, but also reflect the physical and mechanical mechanism of landslide occurrence, and can well handle the complex nonlinear relationship in landslide evaluation, realizing the complementary advantages of various data and models, making up for the defects of traditional single data types or single models, effectively reducing the impact of the decline in classification accuracy due to statistical factor data errors, and improving the assessment accuracy of landslide hazard. In addition, the configuration model method of this study overcomes the problem of slow or even stagnant accuracy improvement when the physical and mechanical model is coupled with the neural network model, and to a certain extent explains the "black box" decision-making mechanism in deep learning, thereby improving the interpretability of the model.

[0020] (3) The research results show that the evaluation accuracy of the Scoops3D-CNN coupling model (the AUC value of the ROC curve is 0.85) is higher than that of the single Scoops3D model (0.78) and the CNN model (0.82). Compared with the traditional single evaluation method, the coupling model can more accurately evaluate the hazard of landslides in the study area. This invention provides a new way to quantitatively evaluate the hazard of regional landslides. The relevant research results can provide quantitative basis and support for landslide disaster prevention and mitigation and active risk control in the study area. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 The present invention is a flow chart of a landslide hazard assessment method based on a convolutional neural network coupling model.

[0022] Figure 2 Schematic diagram of the Scoops3D-CNN coupling model principle and process.

[0023] Figure 3 Performance diagram for Scoops3D physical and mechanical model evaluation.

[0024] Figure 4 Performance graph for CNN convolutional neural network model evaluation based on statistics.

[0025] Figure 5 Performance diagram for the Scoops3D-CNN coupled model evaluation based on both physics and data. DETAILED DESCRIPTION

[0026] The specific implementation modes of the present invention are described below so that those skilled in the art can understand the present invention. However, it should be clear that the present invention is not limited to the scope of the specific implementation modes. For those of ordinary skill in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the attached claims, these changes are obvious, and all inventions and creations utilizing the concept of the present invention are protected.

[0027] like Figure 1 As shown, in one embodiment of the present invention, a landslide hazard assessment method based on a convolutional neural network coupling model comprises the following steps: S1. Collect environmental background data of the study area and obtain key statistical factors that affect the formation, development and occurrence of landslides; S2. The stability of the slope in the study area is calculated by Scoops3D physical and mechanical model, and the calculated landslide stability coefficient is used as a physical model factor to establish a landslide hazard assessment factor set with statistical factors; S3. Determine the landslide samples and non-landslide samples, as well as the training set and test set of each sample based on the landslide hazard assessment factor set and the historical landslide distribution data of the study area; S4, inputting the training set of each sample into the convolutional neural network model to obtain a trained coupled convolutional neural network model; S5. Input the test set into the trained coupled convolutional neural network model to obtain the landslide hazard classification evaluation.

[0028] like Figure 2 As shown in this embodiment, in order to comprehensively consider various evaluation factors affecting the hazard of landslides and give full play to the advantages of various evaluation models, a Scoops3D-CNN coupling model is proposed, which includes a convolutional neural network model driven by a comprehensive Scoops3D physical model and statistical data, and is used for quantitative hazard evaluation of landslides. In this coupling model, the present invention uses the landslide stability coefficient calculated based on the Scoops3D physical model and various main factors affecting the formation, gestation and occurrence of landslides as landslide hazard evaluation factors, inputs them into the convolutional neural network, and strictly controls the main component proportions of physical and mechanical factors and other statistical evaluation factors.

[0029] In S1, the key statistical factors include topographic factors, geological factors, meteorological and hydrological factors, and seismic factors, among which the topographic factors include slope, aspect, and relative height difference; the geological factors include engineering geological rock groups and the distance from faults; the meteorological and hydrological factors include the distance from the water system and the average maximum daily rainfall over many years; and the seismic factor is the earthquake intensity.

[0030] In this embodiment, based on the collection of environmental background data on topography, geology, meteorology, hydrology, earthquakes, etc. in the study area, the main influencing factors affecting the formation, gestation and occurrence of landslides are obtained and used as statistical factors, with a total of 8 factors in 4 categories.

[0031] In S2, the landslide stability coefficient is calculated F s The specific expression is:

[0032] In the formula, i Number the rows of the 3D cylinder, j is the column number for the 3D cylinder, R ij is the search radius, e ij To search for the distance from the center of the sphere to the center of mass of the three-dimensional cylinder, A ij is the sliding surface area of ​​the potential sliding body, c ij is the cohesion of the potential sliding body, φ ij is the internal friction angle of the potential sliding body, α ij is the apparent inclination angle of the potential sliding body, W ij is the gravity of the potential sliding body, u ij is the pore water pressure, k ij is the horizontal vibration load.

[0033] In this embodiment, the physical and mechanical parameters and model parameters of the slope rock and soil in the study area are determined by combining indoor and outdoor tests, engineering geological analogy and literature data. The stability of the slope in the study area is calculated based on the Scoops3D physical and mechanical model to obtain the landslide stability coefficient of the slope in the study area.

[0034] The obtained landslide stability coefficient is used as the physical model factor of the physical and mechanical mechanism, and together with the above 8 factors, it constitutes a landslide hazard assessment factor set of 9 factors in 5 categories.

[0035] The present invention adopts the same ratio of landslide and non-landslide samples in sample selection, and the ratio of training set to test set is 7:3. In order to extract the advantages of physical model factors and statistical factors to the greatest extent, each feature is input separately without interfering with each other.

[0036] In S4, the convolutional neural network model includes a first convolutional layer, a first pooling layer, a second convolutional layer, a second pooling layer, a third convolutional layer, a third pooling layer, a fully connected layer and an output layer which are connected in sequence.

[0037] In S4, the method for obtaining the trained convolutional neural network coupling model is specifically as follows: S41, inputting each factor in the training set into the convolutional neural network model respectively, and specifying the weight ratio of the physical model factor and the statistical factor by the principal component proportion method, thereby controlling the calculation process of the convolutional neural network model; S42. Use the cross entropy loss function to calculate the error between the prediction result and the true label, and use the Adam optimizer to train the model to minimize the prediction error and obtain the trained convolutional neural network coupling model.

[0038] In the S41, the physical model factors and statistical factors each account for 50% of the principal components affecting the convolutional neural network model calculation. In this embodiment, the input data is processed using the principal component proportion method, which specifies the weight ratio of the physical model factors and statistical factors in the final coupling model calculation. In this coupling model, the physical model factors account for 50% of the principal components affecting the convolutional neural network calculation (PCA), while the statistical factors account for the remaining 50%. This coupling method has not appeared in other landslide hazard assessment methods based on machine learning. The present invention proposes this innovative coupling strategy for the first time. In this way, the coupling model can comprehensively consider the effects of multiple influencing factors such as geology, geomorphology, meteorology and hydrology on the formation, breeding and occurrence of landslides, and can also effectively reflect the physical and mechanical mechanisms of landslide occurrence. This method can not only handle the complex nonlinear relationship in landslide evaluation, but also realize the complementary advantages of the physical model and the convolutional neural network model, making up for the shortcomings of the traditional single model in that it cannot simultaneously consider multiple influencing factors and physical and mechanical mechanism processes. In the S42, the cross entropy loss function L The specific expression is:

[0039] In the formula, N is the total number of samples in the dataset, y i For the i The true labels of samples, For the i The predicted output of samples; The specific expression of Adam optimizer is:

[0040] In the formula, is the updated model weight parameter, is the model weight parameter before updating, η is the learning rate, λ is the L2 regularization coefficient, is the sign of the partial derivative, w is the convolution kernel.

[0041] In this embodiment, in order to evaluate the effect of the coupling model adopted by the present invention in the landslide hazard classification evaluation, the principal component (PCA) values ​​of each factor affecting the accuracy of the model were first analyzed and discussed. Subsequently, the evaluation performance and results of the coupling model were compared with those of a single Scoops3D physical mechanics model and a single convolutional neural network (CNN) artificial intelligence model based on statistical influencing factors. In the landslide hazard evaluation coupling model proposed in the present invention, the proportion of the PCA principal components of the Scoops3D physical model evaluation factor Fs and other statistical factors in the coupling model is a very important factor affecting the accuracy of the coupling model. In order to analyze the evaluation effect of each statistical factor PCA on the coupling model, the evaluation performance of the coupling model under different proportions of the average values ​​of a large number of 2-type factor PCA was calculated, and the specific results are shown in Table 1.

[0042] Table 1 Evaluation reliability of coupling model under different proportions of 2-factor PCA

[0043] As can be seen from Table 1, when the PCA proportion of the physical model factor is 50% or 70%, the evaluation accuracy of the coupling model reaches the highest, and the area under the ROC curve (AUC value) is 0.86. As the PCA proportion of the physical model evaluation factor further increases or decreases, the evaluation accuracy of the coupling model shows a downward trend. This shows that the physical model factors and other statistical factors play an important role in landslide hazard assessment. Based on the above research, a coupling model in which the PCA proportion of the physical model factor and other statistical factors is 50% each was selected, and the evaluation performance was compared with the Scoops3D physical mechanics model and the CNN artificial intelligence model based solely on statistical evaluation factors. The results are shown in the figure. Figures 3-5 shown.

[0044] Depend on Figures 3-5 It can be seen that the evaluation accuracy of the coupled model is higher than that of any single model, and it is the highest among the three models, with an AUC value of 0.85; followed by the CNN artificial intelligence model based on statistical evaluation factors, with an AUC value of 0.82, and the Scoops3D physical mechanics model with a relatively low evaluation accuracy, with an AUC value of 0.78. This study shows that when deep learning and physics-based models are combined, the model performance can be improved to varying degrees.

[0045] The comparison of the evaluation results of the coupling model and each single model is shown in Table 2. As can be seen from Table 2, the total area of ​​high and relatively high hazard areas in the coupling model accounts for 12.1%, and the total landslide point density and surface density are 58.8% and 57.8% respectively. Compared with the Scoops3D model (the total area of ​​high and relatively high hazard areas accounts for 15.1%, and the total landslide point density and surface density are 51.6% and 40.7% respectively) and the CNN model based on statistical data (the total area of ​​high and relatively high hazard areas accounts for 16.4%, and the total landslide point density and surface density are 57.6% and 52.5% respectively), the coupled model has higher point density and surface density of landslides in the relatively small high and extremely high hazard areas than the two single models. This shows that the hazard assessment results of the coupling model are more consistent with the distribution of actual historical landslides, and its results are more in line with reality.

[0046] Table 2 Comparison of evaluation results of three types of models

[0047] According to the model accuracy evaluation, it is clear that the coupled model performs well in regional landslide hazard classification, showing significant improvement compared with the single convolutional neural network model and the physical mechanics model. This shows that the coupled model can provide more objective assessment results than the single convolutional neural network model or the single physical mechanics model, and its results have higher reliability and accuracy. When using a single data-driven model for landslide susceptibility assessment, it is still challenging to achieve high-accuracy landslide prediction at a regional scale due to computational complexity. Although the Scoops3D physical and mechanical model takes into account the physical and mechanical mechanisms and the three-dimensional geometric characteristics of the slope body, it requires detailed physical and mechanical parameters of the rock and soil body. The uncertainty of these parameters directly affects the prediction accuracy of the coupled model. In addition to being unable to consider the physical and mechanical mechanisms, the accuracy of the model driven solely by statistical data is significantly affected by the quality of statistical data. The occurrence of landslides is usually caused by the combined effects of internal basic factors and external driving forces. By combining the advantages of physical models and data-driven models, the accuracy and interpretability of landslide susceptibility assessment can be improved from a more comprehensive perspective.

[0048] The beneficial effects of the present invention are as follows: the present invention provides a landslide hazard assessment method based on a convolutional neural network coupling model. On the basis of investigating, remote sensing interpretation and analyzing the environmental background conditions and disaster characteristics of historical landslide disasters in the study area, the Scoops3D-CNN coupling model is innovatively proposed for quantitative assessment of the hazard of landslides in the study area. In the coupling model, the weight ratio of the physical model evaluation factor and the statistical evaluation factor is adjusted for the first time by using the principal component ratio. Compared with the method of direct classification using the convolutional neural network, this method can find the configuration model method that is most optimized and most in line with actual research. The coupling model proposed by the present invention can not only comprehensively consider the influence of various influencing factors (geology, geomorphology, meteorology and hydrology, etc.) on the incubation and occurrence of landslides, but also reflect the physical and mechanical mechanism of landslide occurrence, and can well handle the complex nonlinear relationship in landslide evaluation, realize the complementary advantages of different data and models, make up for the defects of traditional single data types or single models, effectively reduce the influence of the decline in classification accuracy due to statistical factor data errors, and improve the assessment accuracy of landslide hazard. In addition, the configuration model method of this study overcomes the problem of slow or even stagnant accuracy improvement when the physical mechanics model is coupled with the neural network model, and to a certain extent explains the "black box" decision-making mechanism in deep learning, thereby improving the interpretability of the model.

[0049] The research results show that the evaluation accuracy of the Scoops3D-CNN coupling model (the AUC value of the ROC curve is 0.85) is higher than that of the single Scoops3D model (0.78) and the CNN model (0.82). Compared with the traditional single evaluation method, the coupling model can more accurately evaluate the hazard of landslides in the study area. The present invention provides a new way for the quantitative evaluation of regional landslide hazard, and the relevant research results can provide quantitative data support for landslide disaster prevention and mitigation and active risk control in the study area.

[0050] In the description of the present invention, it is necessary to understand that the orientation or positional relationship indicated by the terms "center", "thickness", "upper", "lower", "horizontal", "top", "bottom", "inner", "outer", "radial", etc. is based on the orientation or positional relationship shown in the drawings, and is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present invention. In addition, the terms "first", "second", and "third" are used only for descriptive purposes, and cannot be understood as indicating or implying the relative importance or the number of implicitly specified technical features. Therefore, the features defined by "first", "second", and "third" may explicitly or implicitly include one or more of the features.

Claims

1. A landslide hazard assessment method based on a convolutional neural network coupling model, characterized in that: The following steps are involved: S1. Collect environmental background data of the study area and obtain key statistical factors that affect the formation, development and occurrence of landslides; S2. The stability of the slope in the study area is calculated by Scoops3D physical and mechanical model, and the calculated landslide stability coefficient is used as a physical model factor to establish a landslide hazard assessment factor set with statistical factors; S3. Determine the landslide samples and non-landslide samples, as well as the training set and test set of each sample based on the landslide hazard assessment factor set and the historical landslide distribution data of the study area; S4, inputting the training set of each sample into the convolutional neural network model to obtain a trained coupled convolutional neural network model; S5. Input the test set into the trained coupled convolutional neural network model to obtain the landslide hazard classification evaluation.

2. The landslide hazard assessment method based on the convolutional neural network coupling model according to claim 1 is characterized in that: In S1, the key statistical factors include topographic factors, geological factors, meteorological and hydrological factors, and seismic factors, among which the topographic factors include slope, aspect, and relative height difference; the geological factors include engineering geological rock groups and the distance from faults; the meteorological and hydrological factors include the distance from the water system and the average maximum daily rainfall over many years; and the seismic factor is the earthquake intensity.

3. The landslide hazard assessment method based on the convolutional neural network coupling model according to claim 1 is characterized in that: In S2, the landslide stability coefficient is calculated F s The specific expression is: In the formula, i Number the rows of the 3D cylinder, j is the column number for the 3D cylinder, R ij is the search radius, e ij To search for the distance from the center of the sphere to the center of mass of the three-dimensional cylinder, A ij is the sliding surface area of ​​the potential sliding body, c ij is the cohesion of the potential sliding body, φ ij is the internal friction angle of the potential sliding body, α ij is the apparent inclination angle of the potential sliding body, W ij is the gravity of the potential sliding body, u ij is the pore water pressure, k ij is the horizontal vibration load.

4. The landslide hazard assessment method based on the convolutional neural network coupling model according to claim 1 is characterized in that: In S4, the convolutional neural network model includes a first convolutional layer, a first pooling layer, a second convolutional layer, a second pooling layer, a third convolutional layer, a third pooling layer, a fully connected layer and an output layer which are connected in sequence.

5. The landslide hazard assessment method based on the convolutional neural network coupling model according to claim 4 is characterized in that: In S4, the method for obtaining the trained convolutional neural network coupling model is specifically as follows: S41, inputting each factor in the training set into the convolutional neural network model respectively, and specifying the weight ratio of the physical model factor and the statistical factor by the principal component proportion method, thereby controlling the calculation process of the convolutional neural network model; S42. Use the cross entropy loss function to calculate the error between the prediction result and the true label, and use the Adam optimizer to train the model to minimize the prediction error and obtain the trained convolutional neural network coupling model.

6. The landslide hazard assessment method based on the convolutional neural network coupling model according to claim 5 is characterized in that: In S41, the physical model factors account for 50% of the principal components affecting the convolutional neural network model calculation, and the statistical factors account for 50% of the principal components affecting the convolutional neural network model calculation.

7. The landslide hazard assessment method based on the convolutional neural network coupling model according to claim 5 is characterized in that: In S42, the cross entropy loss function L The specific expression is: In the formula, N is the total number of samples in the dataset, y i For the i The true labels of samples, For the i The predicted output of samples; The specific expression of Adam optimizer is: In the formula, is the updated model weight parameter, is the model weight parameter before updating, η is the learning rate, λ is the L2 regularization coefficient, is the sign of the partial derivative, w is the convolution kernel.

Images