Polarimetric Roll-Invariant Features and Applications for Polarimetric Synthetic Aperture Radar Ship Detection: A comprehensive summary and investigation

Abstract

Polarimetric radar, which can acquire full polarization information, has become mainstream in microwave remote sensing. However, radar target scattering responses are strongly orientation dependent, which makes applications such as target detection and recognition more difficult. Polarimetric roll-invariant features, which are independent of target orientations along the radar line of sight (LOS), exhibit great importance and have achieved successful applications in many fields. During the development of radar polarimetry, a number of polarimetric roll-invariant features have been found and reported. A comprehensive summary and investigation of these polarimetric roll-invariant features is necessary and valuable for current studies and further developments. This article is dedicated to this purpose and contains two main contributions. First, according to the authors’ knowledge, polarimetric roll-invariant features are thoroughly investigated. Based on their original derivation fashion, they are partitioned into six groups, including those derived from the Sinclair matrix, the Graves matrix, the polarimetric coherency and covariance matrices, eigenvalue–eigenvector-based decompositions, polarimetric coherence/correlation patterns, and the Kennaugh matrix. Meanwhile, their expressions, inner relationships, and physical interpretations are further examined and presented. Second, since ship detection is an important application for polarimetric synthetic aperture radar (PolSAR), the potential of these independent features for PolSAR ship detection is investigated and demonstrated. Experimental studies with Radar Satellite-2 ( RADARSAT-2 ) and Gaofen-3 PolSAR datasets are carried out. It is validated that the selected polarimetric roll-invariant features exhibit superior performance compared with normal polarimetric features.