A landslide susceptibility evaluation method based on deep learning

Abstract

The invention discloses a landslide susceptibility evaluation method based on deep learning. The layers are respectively input to different channels of the convolutional neural network CNN, and the extracted features are weighted and fused by the multi-channel weighted convolution module, and the layer adaptive classification convolution is performed after classification by the layer number adaptive network module using the information entropy clustering method. Output the stacked feature map, input the fully connected layer to obtain the integrated feature map, use the Softmax layer to perform probability regression on the integrated feature map, and output the landslide susceptibility evaluation results. The method uses the information gain ratio to perform multi-channel weighting on the features extracted from each factor layer to improve the convergence speed of the model training. The convolution operation of the adaptive layer number is carried out for the classification of influencing factors with different information richness, and the optimal number of layers is solved. question.

Description

technical field [0001] The invention relates to the technical field of geological disaster risk analysis, and more particularly to a landslide susceptibility evaluation method based on deep learning. Background technique [0002] Landslide susceptibility evaluation is to comprehensively analyze various geological environmental factors, historical landslide data, physical laws of landslides and other elements in the study area to determine the probability of future landslides in the study area. At present, landslide susceptibility mapping methods are mainly divided into empirical models (expert system scoring method (Zhang Wen et al., 2010), AHP (Lyu et al., 2018)), statistical models (frequency ratio method (Li et al. , 2016; Zhang et al., 2020), the amount of information method (Wang Xiaohao, 2020), the right of evidence method (Guo Zizheng et al., 2019)), machine learning methods (Ermini et al., 2005; Sun et al., 2021; Tan Long et al., 2014) and an integrated approach of ...

AI Technical Summary

Problems solved by technology

[0003] Most studies are based on the single pixel value of each factor layer where the landslide point is located, ignoring that the occurrence of the landslide is also related to its surrounding environment, and there is a certain error in using a single pixel to represent the location of the entire landslide, and the CNN model cannot be fully utilized Advantages in spatial data processing

Images