Rapid assessment of landslide risk level based on deep learning

Abstract

A rapid assessment of landslide risk level is carried out based on deep learning to speed up the process of determining landslide risk level. The problems of current assessment methods include cumbersome data processing, and the time-consuming and labor-intensive evaluation of risk factors. In this paper, 14 risk factors were analyzed, including slope topography, geology, and hydrology. The analysis involves using the stacked autoencoder (SAE) for training and testing based on an error backpropagation algorithm, and the data collected from engineered slopes as samples. Through multiple times of training, an optimized SAE model was obtained. The landslide risk level can be quickly and accurately determined by the SAE model. Then, the generalization tests further illustrate this. The complicated and lengthy data pre-processing process required in the traditional method was eliminated. This enables the risk level to be quickly determined by SAE after the landslide risk factors are identified. Therefore, a lot of time and cost will be saved by applying the proposed SAE to carry out landslide risk assessment. This will be a new direction of engineering risk assessment.