A physical statistical clutter model for active sonar scenarios with variable signal-to-noise ratios

Abstract

Active sonar classification algorithms need to be robust in preventing operator overload while not being misled by false targets. This talk describes a new 3-parameter statistical sonar clutter model that not only provides a physical context for relating the characteristics of normalized matched-filter echo-data distributions to scatterer attributes, but scatterer information that is largely independent of its peak signal-to-noise ratio (SNR) value. It extends our 2-parameter Poisson-Rayleigh model (Fialkowski and Gauss, IEEE JOE, 2010) by adding a quantitative measure of scatterer spatial dispersion to its measures of scatterer density and relative strength. Maximum likelihood estimates of the clutter model’s 3 parameters were derived from mid-frequency (1-5 kHz) shallow-water active sonar data containing returns from biologic, geologic and anthropogenic objects with differing spatial and scattering characteristics. The resulting clutter model’s probability density functions not only fit the non-Rayleigh data well while displaying an insensitivity to SNR, but the dispersion parameter values were consistent with the known spatial characteristics of the scatterers and the values’ ping-to-ping variance correlated strongly with clutter object class, all of which are encouraging with regard to developing robust physics-based active classification algorithms. [Work supported by ONR.]