Fast principal component analysis‐based detection of small targets in sea clutter

Abstract

Aiming at fast small target detection in high-resolution maritime surveillance radars, this letter proposes a fast detection method using multiple salient features extracted from radar returns and principal component analysis (PCA) in the feature space. It consists of offline clutter feature subspace selection and online fast decision. The PCA of the training feature vectors of sea clutter is computed to generate the recombined features for compact representation of sea clutter features, and the training feature vectors of simulated target returns plus sea clutter are used to find the optimal clutter feature subspace spanned by significant recombined features. The distance of a feature vector under test from the optimal clutter feature subspace is used as the test statistic for fast online decision. Experimental results on the recognised and open IPIX and CSIR radar databases show that the proposed detector reduces decision time up to two scales of magnitude and keeps competitive performance in comparison with the existing feature-based detectors.