Advancement of risk assessment method for slope failures using deep neural network

Abstract

  This study proposes a method for the identification of locations for which risk of slope failure during heavy rain is high. Using teaching data created from areas of slope failure during heavy rain, a prediction model is created to assess slope failure risk for individual detailed digital elevation models (DEMs). This prediction model achieves a12％ improvement in accuracy over the conventional technique. Appropriate assessments are also conducted for regions in which prediction has been difficult in the past. In addition, this study has succeeded in simplifying the structure of the deep neural network (DNN) through the collection of high-quality teaching data. Based on this method of analysis, a method of assessing the degree of risk for mountain streams is proposed.