Ocean Clutter Modeling for Ship Detection

This work addresses the problem of covariance matrix estimation for ocean clutter modeling. For ship detection based on polarimetric synthetic aperture radar (PolSAR) imagery and constant false alarm rate (CFAR) detectors, accurate ocean clutter modeling is essential. The covariance matrix provides all the polarimetric information of the ocean clutter and its estimate is always involved in PolSAR detection [1]. The aim of this work is to investigate and compare the behavior of different covariance matrix estimators, i.e., the sample mean, fixedpoint, and maximum likelihood estimators. An approximate maximum likelihood covariance matrix estimator is also proposed and discussed for better computational efficiency. Their performances are evaluated in terms of the Kullback-Leibler (KL) matrix distance, and computational efficiency. Various textured ocean clutter conditions are considered, ranging from high texture to the non-textured case with Gaussian clutter. Experiments are performed on simulated ocean clutter data.