Remote sensing and Gis
Khalil Valizadeh_Kamran; Fatemeh Adimy; MohammadHosain Pourghorban Benam
Abstract
One of the hazards that threatens the existing infrastructures in different regions is the phenomenon of landslides. The present study tries to identify the prone areas of this natural phenomenon in Khalkhal city, which was done using the neural network method. For this purpose, 9 factors affecting landslides ...
Read More
One of the hazards that threatens the existing infrastructures in different regions is the phenomenon of landslides. The present study tries to identify the prone areas of this natural phenomenon in Khalkhal city, which was done using the neural network method. For this purpose, 9 factors affecting landslides were identified and prepared, the layer of landslides that happened, obtained from aerial photos and satellite images and field visits, and using non-sliding points in the region, perceptron neural network training data. They created several layers. In order to model the neural network, these data were transferred to the MATLAB 2016 software after initial preparation in the ARC GIS 10.5 software environment and were trained using MLP neural network coding to deal with the data that they have not encountered. , make predictions. The structure of the designed neural network was selected from among the many networks that were created and tested, 1-12-9, which obtained 9 inputs as the number of effective criteria, 12 neurons in the middle layer and one neuron and layer for the output of the network. The results of the validation chart of the neural network (ROC) model show a high accuracy of 95% of the created model in predicting sliding pixels. According to the results, 0.57%, 0.11%, 0.07%, 0.06% and 0.07% of the studied area were placed in very high, high, medium, low and very low classes, respectively. Due to the fact that the largest area of the region is in a very high risk class,
