UNSUPERVISED CLASSIFICATION METHOD FOR HYBRID POLARIMETRIC SAR IMAGES BASED ON SCATTERING MECHANISMS AND WISHART CLASSIFIER

摘要

In the past decade, a new polarimetric synthetic aperture radar (PolSAR) technique, the compact polarimetry (CP), which requires transmitting onlyrnone particular polarization, instead of two as in full polarimetry (FP), has rose wide attentions for its main advantages such as wider swath coverage (compared with FPrnSAR), simple hardware requirement and unique self-calibration feature. In this paper, we are focused on the unsupervised classification application of the so-calledrnhybrid PolSAR, which is one manifestation of the CP SAR. In our proposed method, both polarimetric features extracted from the roll-invariant m-chi decompositionrnalgorithm and the Wishart classifier are utilized. First, the initial scattering categories are derived from the m-chi classification algorithm. Then, pixels with no clearlyrndefined dominant scattering mechanism are excluded, and the resulting categories are expanded into a specific number of classes. These derived classes are taken asrntraining samples of the Wishart classifier. At last, the effectiveness of the proposed algorithm is validated with AIRSAR and RADARSAT-2 datasets over San Francisco.