Neural network techniques applied to seismic event classification

Abstract

An artificial neural network is incorporated as part of a software simulation system for the purpose of classifying seismic events from waveform data. Unprocessed seismograms are not well suited for presentation to neural networks because of the large number of data points required to represent a seismic event in the time domain. Parametric representation of the seismic event provides adequate information for accurate event classification, while significantly reducing the minimum size and is comprised of five signal classes, with 2400 samples per seismic trace. Each waveform in this database is parametrically represented by ten central moments. These moments are presented to the neural network for classification. Correct seismic event classification accuracy exceeds 98%.