Polarimetric SAR Target Decomposition based on sparse NMF

Abstract

Polarimetric Targets Decomposition is to decompose polarimetric data into several components by different physical scattering mechanism. Traditional decomposition methods are model-based, whose basis matrix is fixed to typical physical scatterers, such as surface, double bounce, volume. However, those methods cannot distinguish the differences of two kinds of surface scattering, or two kinds of double bounce scattering. On the other hand, if there were only double bounce scatterers in the image, those methods still decomposed it into several fixed basis physical scattering mechanism. This paper used Non-Negative Matrix Factorization (NMF) to do polarimetric targets decomposition. NMF is a part-based technique that have been frequently applied on computer vision, face recognition, etc‥ This paper projects coherency matrix on Pauli basis space, and then applies NMF to factorize probability distribution matrix. Results show NMF based polarimetric decomposition can divide surface scatterers into two parts — sea surface and ground surface, and the basis matrix always represents those kinds of components that can describe the mainly physical mechanism of the image being decomposed best.