The Extended Polarimetric Whitening Filter and Its Application to Target Detection in Polarimetric Synthetic Aperture Radar Images

Abstract

In this letter, we propose the extended polarimetric whitening filter (EPWF) for target detection in polarimetric synthetic aperture radar (PolSAR) images. We adopt finite mixture models fitted by the expectation-maximization algorithm for modeling PolSAR clutter. Specifically for single-look and multilook PolSAR data, the Gaussian mixture model (GMM) and Wishart mixture model (WMM) are applied, respectively. At the core of the EPWF, we present the closed-form solution to the optimal weighting matrix that minimizes the speckle of filtered clutter. We also derive the null hypothesis distribution of the EPWF detector to determine the detection threshold given a false alarm rate. The effectiveness of the EPWF is evaluated on both simulated and real PolSAR data. Experimental results demonstrate that the GMM and WMM are suitable for PolSAR clutter with channel-dependent texture. Compared with alternative detection statistics, the speckle level of the EPWF is lower in terms of the standard-deviation-to-mean ratio, which leads to a preferable target detection performance of the EPWF detector.