Application of a minimum probability of error classifier with Linear Time-Varying pre-filters for buried target recognition

Abstract

In this paper we overview the theory of Linear TimeVarying (LTV) filters and investigate via simulation their application to buried target classification in challenging nonstationary environments; in particular, environments where noise is not only nonstationary but exhibits statistical properties that are not known a priori. We then propose an extension of the Minimum Probability of Error (MPE) classifier (a/k/a Minimum Distance Receiver) by pre-processing the received data through a bank of LTV filters before the calculation of each test statistic via the MPE classifier. The proposed augmented MPE classifier is shown to outperform the conventional MPE classifier via simulation.