CFAR detection of extended objects in high-resolution SAR images

This paper deals with the problem of constant false alarm rate detection of extended objects in high-resolution synthetic aperture radar (SAR) images. Starting from the receiver proposed by Guida and Longo, where some relevant properties of the location-scale distributions are exploited for ensuring constant false alarm against Weibull background, the authors consider the application of the algorithm to the case of spotlight SAR images collected by an airborne sensor developed by the Defense Evaluation and Research Agency laboratories. The algorithm has been reconceived for specific application to high-resolution SAR images where the reference window is order of magnitudes larger than in the usual radar case. As a first step, a careful analysis based on some real images has been made for testing the validity of the Weibull distribution hypothesis. Results are then presented for different values of the system parameters (censoring depth and size of the reference window) and demonstrate that the proposed detector is able to operate under changeable conditions of the underlying background (vegetation and asphalt mixed with grass) and of the object size and shape. A little care, however, should be taken to avoid overload situations that occur when the object dimensions become comparable with the size of the reference window. Finally, real-time applications may be foreseen, since the computational charge of the processor is generally low and increases weakly with the number of samples used for parameter estimation.