Research On Polarimetric SAR Signature Analysis And Classification

Polarimetric synthetic aperture radar (PolSAR) is an advanced instrument for obtaining the information of remote sensing. By measuring radar echoes of various combinations of transmitting and receiving polarizations from scattering media, PolSAR provides more abundant back-scattering information of targets than the single-polarization SAR. It enhances the ability for remote sensing the Earth’s environment, such as hazard monitoring, target detection, resource survey, city planning, crop estimation and forest sensing, as well as military surveillance, etc. Therefore, it is a hot topic that how to extract useful information from abundant polarimetric SAR data for interpreting and analyzing the land cover.Firstly, some basics of polarized electromagnetic waves are introduced, such as the polarization state, the description of polarimetric vector, the conversion of polarization basis and the polarimetric representation in terms of power, etc. This sets up the stage for understanding polarimetric scattering mechanisms.Then, we investigate the algebra operation among different polarimetric channels and explain how to synthesize the polarimetric response at different polarization basis. Some canonical scattering mechanisms as well as the polarimetric decomposition theorems are introduced. It not only reveals the connection between polarimetric signatures and scattering mechanisms, but also clarifies the physical meaning of each polarimetric signature. Further, we utilize the polarimetric signatures derived from H/a method for land cover classification. Based on the Bayesian maximum likelihood criterion, the unsupervised Wishart method is employed to obtain a better classification result. The experimental results on a real data show the validity.Various polarimetric signatures can be obtained by target decomposition techniques, which are of great help for characterizing the land cover. It is straightforward to combine these polarimetric signatures together and formulate them as a three-mode tensor. On the basis of multi-signatures combination, we propose a novel polarimetric feature dimensionality reduction scheme that can be used to extract the intrinsic feature set. Then, the feature set is fed to a classifier to perform classification. Compared with the conventional matrix-based dimensionality reduction methods, the proposed method improves the classification accuracy. The results on simulated data and real data show the effectiveness of the proposed method.