Classification of active sonar detections with class specific method

Abstract

The use of the Class Specific Method (CSM) [P. M. Baggenstoss, Trans. Sig. Proc. 47, 3428–3432 (1999)] is demonstrated with numerically simulated data. The two models of a cylindrical shell and a fish school are shown to generate sufficient statistics with appropriate assumptions. The common reference signal for the CSM is Gaussian white noise. An interpretation of the resulting conditional probability density functions is given, which leads to a numerical regularization. The resulting implementation of the CSM is demonstrated using numerically modeled data of variable bandwidth with variable amounts of additive Gaussian white noise. This approach enables the evaluation of bandwidth requirements for sufficient classification performance. The method is shown to be generalizable to include effects of propagation. [Work supported by the Office of Naval Research.]