ACOUSTIC SIGNAL CLASSIFICATION USING UNCONVENTIONAL TECHNIQUES

Advances in active acoustic systems exacerbate the problem of signal classification in a high contact-density environment. Increased contact loading forces the operator to classify contacts more rapidly, reducing his effectiveness. The key to early classification is an understanding of target response in terms of vibrational mode and localization of dominant contributors. Phase I will develop this understanding of target response and permit synthesis of classification methods utilizing unconventional signal processing techniques. These techniques, inconjunction with visually-based image processing techniques, will form an integrated detection and feature extraction process that promises improved detection and extraction of low-SNR features (such as modal response) contained within beam-formed, time-frequency data. We seek to understand the key signature features revealed by this analysis approach, with a focus toward automation of the process. Phase I will evaluate the feature extraction and classification capabilities of the unconventional techniques, which include wavelet transforms, higher-order spectral analysis, time-frequency distributions, and image processing, on simulated acoustic data. We will also research the applicability of advanced GUI methods to user displays and controls. Phase II will implement a Combined Acoustic Signature Analysis Station for validation of the most promising techniques against real measured data.