Application of Signal Analysis and Pattern Recognition to Study a Simple Ground Motion Problem

摘要

A simple problem involving the identification of an explosive source as being unbermed or bermed using a pattern recognition based analysis of buried ground accelerometer measurements is presented. This problem illustrates the advantages of computerized information extraction from the measured waveforms. Information was extracted from the frequency and cepstrum descriptions of the waveforms in addition to the more traditional time domain information. These signal features were incorporated into a Fisher's Linear Discriminant pattern recognition procedure. Previously unseen signals were classified with up to 100% accuracy depending on which features were used. Close in explosive source measurements present unique problems to a pattern recognition based analysis approach. These problems are reviewed and approaches illustrated.