Polarimetric SAR Image Classification Using Geodesic Distances and Composite Kernels

The covariance/coherence matrices are the mostcommon way of representing polarimetric information in the polarimetric synthetic aperture radar (PoISAR) data andhave been extensively used in PoISAR classification. Since PoISAR covariance and coherence matrices are Hermitian positive-definite, theyform a nonlinear manifold, rather than Euclidean space. Though the geodesic distance measures defined on a manifold are suitable for describing similarities of PoISAR matrix data, the nonlinearity of themanifold oftenmakes the involved optimization problems awkward. To address this problem, we propose to embed the manifold-based PoISAR data into a high (infinite)-dimensional reproducing kernel Hilbert space by Stein kernel and log-Euclidean kernel. Besides, we introduce the composite kernel into the sparse representation classification in order to exploit the spatial context information of PoISAR data. The proposed method is assessed using different PoISAR datasets. Experimental results demonstrate the superior performance compared with the methods without the use of contextual information.