Detection of targets from electro-optical and SAR data using chaotic predictors and optimal CFAR detectors

This paper presents a new clutter suppression and target detection technique for locating small land targets in SAR and electro-optical images. The method is based on adaptive nonlinear (chaotic) predictors which provide an estimate of the local SAR and electro-optical clutter and uses this estimate to suppress the clutter and hence increase the signal to clutter power. The residual errors between clutter estimates and actual data correspond to either clutter noise or target signatures. Hence, dedicated models are employed for clutter noise and targets. The modelling is performed using generalized statistical probability density functions. Statistically optimal Constant False Alarm Rate (CFAR) detectors are used to separate the targets from the residual noise. Although, some similarities may be found in the detection process, when multispectral and multipolarized SAR data are employed, the clutter and target statistics are quite different. The paper presents detection approaches that extend those presented in the literature by including threshold statistics, clutter suppression techniques and generalized statistics. Finally, the issue of detection fusion from dissimilar sensors is addressed. For CFAR detection the optimal Order Statistics, Cell averager, Optimal Weibull, Optimal K-pdf and Optimal Generalized Gamma CFAR detectors are employed. The last three have been developed at A.U.G. Signals Ltd.