Accession Number : ADA320924

Title :   Acoustic Feature Extraction for a Neural Network Classifier.

Descriptive Note : Interim rept. Sep 95-May 96,

Corporate Author : ARMY RESEARCH LAB ADELPHI MD

Personal Author(s) : Wellman, Mark C. ; Srour, Nassy ; Hillis, David B.

PDF Url : ADA320924

Report Date : JAN 1997

Pagination or Media Count : 25

Abstract : Artificial neural networks can perform reliable classification of ground vehicles based solely on their acoustic signatures, if robust features can be identified. We present feature extraction and classification results using simple power spectrum estimates, harmonic line association, and principal component analysis. Algorithm implementation and performance analysis of each feature extraction method are discussed. Also given are preliminary evaluation results of a VLSI (very-large-scale integration) device dedicated to neural network implementation.

Descriptors :   *NEURAL NETS, *ACOUSTIC SIGNATURES, *TARGET CLASSIFICATION, ALGORITHMS, COMPUTER ARCHITECTURE, VERY LARGE SCALE INTEGRATION, POWER SPECTRA, RELIABILITY, GROUND VEHICLES.

Subject Categories : Acoustic Detection and Detectors

Distribution Statement : APPROVED FOR PUBLIC RELEASE