Multi-aspect discrimination of underwater mine-like object objects using hidden Markov models

The problem of classification of underwater targets involves discrimination between mine-like and non-mine-like objects as well as the characterization of background clutter. To improve performance of a given classifier, usually multiple aspects will be fused together in some fashion. In this work, a Hidden Markov Model (HMM) is used to make the overall decision. The HMM is a very powerful tool for using multiple observations to make a decision, as no decision is made until all the evidence is presented. In the past several years, much attention has been given in the area of automatic speech recognition to using multilayer perceptron (MLP) networks for estimating certain probabilities in the HMM framework. Several approaches are taken to this MLP/HMM idea in this paper and the results are compared. The test results presented are obtained on a wideband acoustic backscattered data set collected using four different objects with 1 degree of aspect separation for two different bottom (smooth and rough) conditions.