Unsupervised classification of polarimetric synthetic aperture radar interferometry using polarimetric interferometric similarity parameters and SPAN

Abstract

In this study, a new unsupervised terrain classification method is proposed for L‐band polarimetric synthetic aperture radar (SAR) interferometry (PolInSAR) datasets. It can solve the misclassification problems caused by the inherent scattering characteristics of polarimetric SAR (PolSAR). Compared with PolSAR, PolInSAR is much more sensitive to the position and distribution of orientated scatterers. In this study, the polarimetric interferometric similarity parameter (PISP) is proposed to display this sensitivity. It is defined by three optimised scattering mechanisms obtained from PolInSAR datasets. Therefore, it closely relates to the coherent scattering mechanism, and makes use of the phase and amplitude's information. Combined with the scattering power SPAN, the PISP is used to divide the pixels into initial regions. Then, the optimal coherence set parameters A 1 and A 2 are used to distinguish target types in each region based on the number of independent coherent scattering mechanisms. At last, the clusters are merged based on scattering characteristics. The proposed method not only employs optimum coherence but also uses the information of optimum scattering mechanisms. The effectiveness of the proposed method is demonstrated using German Aerospace Center's (DLR) E‐SAR L‐band PolInSAR datasets.