Automatic Geohazard Detection Using Neural Networks

Abstract

In this paper, we demonstrate the potential of neural networks in the automation of shallow geohazard detection and identification on seismic images. We discuss technical considerations and method limitations. The method used in this paper trains a neural network prediction model to automatically detect features in seismic images by estimating model parameters from a large set of input training images that have been manually interpreted. In this case, the independent variable is the seismic image and the dependent variable is the human interpretation. We used a separate test data set that was not used in training the model to validate our results. The novel approach and workflow presented in this paper is a significant advancement in geohazard detection and identification projects. The time taken to complete such a project using a conventional approach is significantly reduced – our model interprets entire seismic volumes in seconds with consistency, minimal human input and comparable accuracy.