Use of target characteristics for classification

Abstract

Scattering from ribbed cylinders exhibits very significant angular dependence. This angular dependence must be accommodated by any active classifier. A numerical technique for incorporating this dependence was previously reported [Charles F. Gaumond, J. Acoust. Soc. Am. 101, 3130 (1997)]. Using a data set simulating both targets and false targets in a shallow water environment, it is demonstrated that the method can be used as a component of an active classifier and produces an improved probability of detection over energy detection. The technique is based upon physics-based wave packets derived from the backscattered signal of a numerical scattering model. There is only one wave packet, originating from the so-called chalice waves, for the particular frequency band in the data set. These wave packets from a numerical model are then used to obtain features from the experimental data. A probabilistic method is then used to combine the features. The resulting performance is determined through the use of ROC curves. [Work supported by ONR.]