Geohazard risk assessment model based on deep neural network

• This paper establishes a geohazard risk assessment model based on deep neural network, which provides a solution for regional geohazard risk assessment. • The proposed model uses natural indicators for geohazard risk assessment from existing national norms and standards, and then further refined and screened based on geological and natural environment characteristics of study area to finally establish the geohazard indicator system. • With the natural data extracted by remote sensing as the input set and the regional geohazard risk levels officially announced through field exploration and investigation as the output set, the DBN-WOA neural network is used for model training, and finally the landslide risk assessment model is obtained, which overcome the subjective drawbacks of traditional evaluation models. • The assessment results of this model can be objectively verified according to the standards of "double-blind experiment". • Besides accurately describing the current state of geohazard risk in the assessment region, the proposed model can also objectively reflect the historical patterns and future trends of geohazard risk change, as well as the internal driving factors of such changes. In order to improve the accuracy of geohazard risk identification and enhance the control of geological disasters, this paper establishes a geohazard risk assessment model based on deep neural network. In the proposed model, natural indicators for geohazard risk assessment are first selected from existing national norms and standards, and then further refined and screened based on geological and natural environment characteristics of study area to finally establish the geohazard indicator system. Taking landslide risk assessment of Huangu Town as an example, this study area is divided into 647 slope units as evaluation objects by 3D vector telemetric data recognition and automatically dividing technology. Then, taking the natural data extracted by remote sensing as the input set and the regional geohazard risk levels officially announced through field exploration and investigation as the output set, the DBN-WOA neural network is used for model training, and finally the landslide risk assessment model is obtained. At last, according to the "double-blind experiment" standard, the accuracy of this model is verified by the evaluation results of Shuang'an Town. The results show that the proposed approach can accurately describe the current state of geohazard risk in the assessment region. Additionally, it can serve as a reasonable reference for relevant departments to comprehend the current state of the geological environment in the area, predict the trend of changes in the geological environment, and formulate policies and measures.