Order statistic CFAR detectors for speckled area targets in SAR

The authors have extended the standard Goldstein's two-parameter detector, the log T-test (LTT), and developed a robust-LTT (RLTT), using order-statistic estimators for the background moments. The LTT uses maximum likelihood estimates of the local mean and variance, which are replaced by the sample median and a reduced range. In both tests, a pixel-under-test is compared to an adaptive threshold composed of these estimators. The test can be designed to detect both point and area targets. For use with synthetic aperture radar (SAR) images and because of the interest in speckled area targets, the authors have developed a block test, which uses the median of a block of pixels-under-test. To evaluate performance, simulation on a clutter-only SAR image is used. The method is also evaluated in a quantitative way, using a receiver operating characteristics. Results show that the RLTT successfully discriminates area targets against the point targets in clutter background.<<ETX>>