Target Identification Using Wavelet-based Feature Extraction and Neural Network Classifiers

Classification of combat vehicle types based on acoustic and seismic signals remains a challenging task due to temporal and frequency variability that exists in these passively collected vehicle indicators. This paper presents the results of exploiting the wavelet characteristic of projecting signal dynamics to an efficient temporal/scale (i.e. frequency) decomposition and extracting from that process a set of wavelet-based features for classification using a multilayer feedforward neural network for vehicle classification. This effort is part of a larger project aimed at developing an Integrated Vehicle Classification System Using Wavelet / Neural Network Processing of Acoustic/Seismic Emissions on a Windows PC performed under a Phase II SBIR for the US Army TACOM/ARDEC. The data set used for validation consists of ground combat vehicles (e.g. Tanks (T-62, T-72, M-60), Lightweight Utility Vehicle, Tracked APC and Tank Transporter) recorded at the Aberdeen Test Center, MD. Initial results using wavelet-based feature extraction and a feed-forward neural network vehicle classifier employing the Levenberg-Marquardt deterministic optimization learning scheme will be presented.