Active sonar classification using Bayesian decision theory

Abstract

Consideration is given to the performance analysis of optimal sonar classification. To perform active classification, a known waveform is transmitted into a medium and directed toward a region called the test volume. An array of N sensors is used to pick up the backscattered signal energy reflected from the M cells of the test volume, and the data are input into a processing algorithm. The processor is to decide if an object is present and, if so, what kind of object is present. An Eulerian model of each object is developed; that is, each object is characterized by the second-order statistical characteristics of its scattering coefficients. A systematic method for evaluating classifier performance is derived. A sensitivity analysis of processor performance is given and interpreted. An analysis of processor performance versus its angular resolution is described.<>