PolSAR ship detection based on low-rank dictionary learning and sparse representation

To improve the ship detection performance in polarimetric synthetic aperture radar (PolSAR) images under low signal-to-clutter ratio (SCR), this paper presents a new PolSAR ship detection method based on the low-rank dictionary learning and sparse representation. For each pixel, the scattering mechanism information is described via a feature vector formed by the polarimetric covariance matrix elements. The sea clutter pixel feature vectors are assumed to lie in a low-dimensional subspace. The proposed method includes two stages. In the first stage, given a set of noise-corrupted clutter training samples, the low-rank dictionary learning method is used to learn a dictionary. The learned dictionary atoms span the clutter subspace. And each training sample can be represented as the sparse linear combination of the dictionary atoms plus a noise term. In the second stage, the sparse representation of each test pixel feature vector using the learned dictionary is obtained. With the representation coefficient, then a test statistic that does not depend on the scattering intensity is defined. Therefore the proposed method is based on the scattering mechanism and applicable under low SCR. The performance improvement of the proposed method is verified using both the synthetic data and real PolSAR images obtained from a RADARSAT-2 PolSAR data set.