Unsupervised classification of PolInSAR based on improved four-component decomposition

Abstract

In this letter, a new unsupervised terrain classification method is proposed for L-band polarimetric synthetic aperture radar (SAR) interferometry (PolInSAR). It can solve the misclassification problems caused by the inherent scattering characteristics of polarimetric SAR (PolSAR). First, the polarimetric interferometric similarity parameter (PISP) is proposed from PolInSAR data sets. Then, an improved four-component decomposition method is proposed based on PISP. It can partly solve the volume component overestimation problems. Using the improved decomposition result, the image is divided into three basic categories. The optimal coherence set parameters A₁ and A₂ are used to define each category based on the number of independent coherent scattering mechanisms. The proposed method not only employs optimum coherences but also uses the information of optimum scattering mechanisms. The effectiveness of the proposed method is demonstrated using PolInSAR data sets of German Aerospace Center’s (DLR) experimental synthetic aperture radar (E-SAR).