Statistics of synthetic aperture sonar images

We have examined certain statistical aspects of seabed images formed by wideband synthetic aperture sonar. In particular, we have formed high-quality data-adaptive non-parametric kernel estimates from the envelope images. In most cases, the estimated tails of the distribution appear to be substantially heavier than those of the Rayleigh-distribution. Parametric approaches, e.g., fits of the data to a Rayleigh mixture, suggest that the data may be described by a mixture of 2 to 3 Rayleigh populations. Images formed by various degrees of preprocessing are analyzed in this manner. In particular, we examine the dependence on the physical resolution, and on using frequency sub-bands for the image formation.