Sea clutter covariance matrix estimation with data-adaptive selection

In order to remove undesired outliers contained in training data, a new algorithm of sea clutter covariance matrix estimation with data-adaptive selection (DAS) is proposed. Since the emergence of some outliers imposes spectral distribution of sea clutter, the spectral relative entropy is introduced to measure those impacts. A data-adaptive selector employing spectral relative entropy to describe spectral differences between clutter and outliers is designed to reject contaminated training data. Consequently the clutter covariance matrix can be estimated exactly with no scene prior information. Simulation results confirmed the effectiveness of the proposed algorithm.