Active acoustic classification of underwater targets based on scattering resonances.

This paper concerns the neural network based active acoustic classification of underwater targets. The feature vectors utilized in the training of the network are based on the scattering resonance parameters extracted from the echo return. It is expected that these parameters constitute close to an ‘‘optimum’’ set of feature vectors in that they greatly reduce the dimensionality of the echo return while preserving the ‘‘unique’’ aspects related to target identity. The classification performance of a probabilistic neural network classifier is evaluated for several different SNR levels for both monostatic and bistatic scattering configurations. The performance of several different resonance extraction algorithms (including the constrained total least squares technique and the SVD ‘‘Prony’’) is evaluated in varying levels of SNR. The classification study is based on echo returns synthesized from T‐matrix solutions for five elastic targets (three spheres and two finite cylinders).