Landslide Susceptibility Detection Using ResNet

Landslides, one of the most common natural disasters in the world, have caused numerous deaths, infrastructure damage, and monetary losses. In the past, scientists have employed a variety of techniques to predict landslide susceptibility. Unfortunately, there is currently a lack of the most precise algorithm that may be utilized to create a landslide susceptibility model. Therefore, the objective of the current work is to assess how well three distinct machine learning techniques—CNN (Convolutional Neural Network), KNN (K-Nearest neighbor), and ResNet (Residual Neural Network)—can predict landslide susceptibility. In this study, training datasets (70% of the total) and testing datasets (30% of the total) were randomly separated into 19 elements like photos, topographical data, and geological data related to past landslide locations etc. For susceptibility modeling, the following factors were taken into consideration: Additional aspects to take into account include soil texture, lithology, land use/cover, slope aspect, topographic moisture index, proximity to roads, streams, and lineaments, convergence index, profile, and plan curvatures. According to the experimental findings, these two machine learning models accuracy is greater than 0.95. The results show that the suggested technique performs with 98 percent accuracy when using the Resnet-50 model and 96 percent accuracy when using the CNN model. Additionally, it should be highlighted that ResNet performs better than CNN in terms of estimating the likelihood of a landslide occurring, but CNN outperformed less accurate. The suggested ResNet model is reliable and suitable for use in the study area's future hazard reduction and land use planning.