Convolutional Neural Network for First Break Picking in Land Seismic Surveys

Abstract

Summary Current methods of automated first break picking used in seismic data processing software are mostly based on the unsupervised algorithms. Due to the variety of the near-surface structures, the behavior of the signal in the regions of first arrivals may vary from shot to shot. This results in manual parameters selection for different seismic exploration regions. As a result, it takes a lot of time and effort of a seismologist to properly complete this stage. In this work to overcome this problem, we propose a supervised method based on the popular nowadays convolutional neural network. Such approach, however, raises another challenge - CNNs require a large amount of well-labeled data.