Landslide Susceptibility Assessment And Post-earthquake Landslide Identification In Jiuzhaigou Area

Landslide disaster is the most widely distributed and destructive geological disaster in China.Frequent landslides seriously endanger people’s livelihoods and hinder ecological construction.Therefore,pre disaster prevention and post disaster identification are necessary ways to improve people’s livelihoods,enhance their sense of security,and avoid losses caused by disasters.With the advantages of fast response speed and high reliability,the use of remote sensing technology for landslide susceptibility evaluation and landslide recognition has become one of the main methods to solve the above problems.Moreover,the combination of deep learning methods has further improved the efficiency of solving this problem.This paper takes Jiuzhaigou County in Sichuan Province as the research area,and takes the secondary landslide disaster caused by the August 8 earthquake in 2017 as an example.Firstly,a coupled Analytic Hierarchy Process(AHP)neural network model was proposed to evaluate the susceptibility of landslides before disasters.Subsequently,the identification of secondary landslides after disasters was carried out,using Yolov5 neural network and multiscale image segmentation based methods for landslide identification and quantitative analysis.The main research content of the paper is summarized as follows:(1)Evaluation of landslide susceptibility based on coupled Analytic Hierarchy Process(AHP)-BP neural network model.On the basis of analyzing and summarizing the applicability of empirical model method and machine model method,a coupled analytic hierarchy process BP neural network model is established to evaluate the landslide susceptibility of Jiuzhaigou County;Created a dataset for model training,testing,and prediction;We compared the evaluation effectiveness and accuracy of using the Analytic Hierarchy Process(AHP)model,BP neural network model,and Coupled AHP-BP neural network model separately.(2)Landslide recognition based on Yolov5 neural network.This article elaborates on the unique structural design of the Yolov5 neural network,which is superior to other networks;Solved problems in data integration and preprocessing processes such as overlap and misalignment of high-resolution images;Created a dataset for training,testing,and prediction of the Yolov5 neural network,and conducted multiple training with different parameters;Use multiple accuracy metrics to measure the performance of the network and conduct local area recognition tests.(3)Landslide information extraction based on multi-scale image segmentation and fuzzy membership function.Based on multi-scale image segmentation methods,the optimal scale for image segmentation was explored;Select a fuzzy membership function based on the image target feature threshold to extract landslide target information,and calculate the area of landslide affected areas.