Automated Trace Editing and Refraction Event Picking Using Neural Networks

Abstract

Programs have been developed that use a backpropagation neural network to automatically edit noisy seismic traces and pick first break refraction events. This paper shows that neural network-based trace editing and first break picking can achieve 85 to 98 percent agreement with manual methods for seismic data of moderate to good quality. Productivity improvements over current manual editing and picking techniques for 2D seismic data should range between 60 and 90 percent. For 3D seismic data sets efficiency increases of up to 800 percent have been demonstrated in a production processing environment.