| Lvliang City is located in the Loess Plateau of Western Shanxi Province.It has complex geological environment,Extremely undulating terrain,fragile natural geographical environment,strong human engineering activities and frequent landslide disasters,which have seriously affected the economic development of the region and the safety of people’s lives and property.The study on the susceptibility of regional landslide disasters is of great significance to the economic development of LvLiang city,and can provide reference for the future work of urban geological hazard prevention and control planning.Based on the 1:200000 geological data of Shanxi Province,this study sorted out 226 existing landslide disaster data,made a comprehensive analysis of the development characteristics and gestate background of landslide in the study area,then built a comprehensive index model and a convolutional neural network model with the help of MapGIS,ArcGIS and Keras platforms to evaluate the susceptibility of landslide disaster in LvLiang area.The purpose of this paper is to investigate the effectiveness of the convolutional neural network model in the evaluation of landslide susceptibility in Lvliang city,and to obtain the zoning map of landslide susceptibility of Lvliang city.The main work of this paper is as follows:(1)This paper briefly expounds the general situation of regional geological background in Lvliang,and summarizes the development and distribution characteristics of landslide disasters in the study area.Based on the investigation and analysis of the historical landslide disaster in the study area and the previous experience,six impact factors such as topographic slope,landforms,rock and soil mass,active fracture,rainfall and peak acceleration of ground motion are selected as landslide hazard susceptibility,and a regular grid of reasonable size was determined as the mapping unit of the study.(2)The comprehensive index evaluation model of landslide disaster in the study area was constructed.The weight of each hazard factor was obtained by establishing the analytic hierarchy process(AHP)of three levels of indicators,and each factor was graded and assigned.The comprehensive landslide susceptibility index of each grid unit in the study area was calculated by using the comprehensive index model.The natural break point method was used to make the regional map of landslide susceptibility of LvLiang city on the ArcGIS platform.(3)The possibility of using convolutional neural network to analyze landslide susceptibility was discussed.In this study,ENVI was used to pre-process factor map stacking,and AlexNet-2 model was constructed based on Keras platform to evaluate the susceptibility of landslides in Lvliang City.70% of historical landslide data were randomly selected to be input into the model for training,all parameters were optimized through the training process of trial-and-error method,and the remaining 30% of landslide data was used to perform the test in the trained model.After training the results of CNN model,the weight index of landslide susceptibility in each grid was graded by natural breakpoint method,and the landslide susceptibility zoning map of Lvliang City was made on ArcGIS.(4)The results of the two models were compared and verified by Receiver Operating Characteristic curve and Overall Accuracy,to verify the rationality of the method used in this paper.The proportion of landslide prone partition in the results of the comprehensive exponential model and the convolutional neural network model was similar in the whole figure,and the convolution neural network model used in this paper has higher accuracy than the comprehensive index model.Specifically,in terms of Area Under Curve value and Overall Accuracy value,the results of the convolutional neural network model in this paper are 0.024 and 2.47% higher than those of the comprehensive exponential model,respectively. |
PDF Full Text Request