Model validation of the extremely heterogeneous SAR clutter and its CFAR detection

Based on real extremely heterogeneous synthetic aperture radar (SAR) urban clutter data and Cramer-Von Mises distance, the goodness-of-fit tests were done via Rayleigh distribution, Weibull distribution, K distribution and G distribution, respectively. It is shown that the Rayleigh distribution and Weibull distribution may be mismatched and the K distribution may be only partly matched, but the G distribution may model the real data very well. Furthermore, for the CFAR (Constant False Alarm Rate) detection under the G distributed clutter, the curves of the signal-to-noise ratio versus shape parameter and numbers of reference cells were presented to meet the given detection performance, and the reason was also analyzed that the CFAR losses decrease with the increasing of the distribution parameter. Also, an effective CFAR threshold calculating method was proposed at last.