Vehicle Classification Using a Biological Model of Hearing

The Army is interested in using acoustic sensors in the battlefield to perform vehicle tracking and classification using passive arrays of acoustic microphones and seismic sensors. Here, we present a prototype vehicle acoustic signal classification. To analyze acoustic features of the vehicle signal, we adopt biologically motivated feature extraction models. Physiological and psychophysical research have shown that primary auditory cortex performs to the first order a multi-scale decomposition of the incoming auditory spectra, on axes of log-frequency and time. This decomposition, based on the spectra emerging from a realistic model of the cochlea, is then used as a input to a classifier. Different vector quantization (VQ) clustering algorithms are implemented and tested for real world vehicle acoustic signal, such as Learning VQ, TreeStructured VQ and Parallel TSVQ. Experiments on the Acoustic-seismic Classification Identification Data Set (ACIDS) database show that both PTSVQ and LVQ achieve high classification rates. The advantage of using biologically-based representation and classification algorithms include noise-robustness and existing low-power aVLSI implementations. We present classication results and performance levels. The VQ schemes presented here have the advantage of not having to choose explicitly the features that distinguish one target from another. The burden is shifted to having to choose the ``best’’ representation for the classifier.