Urban land cover change types identification using fully polarimetric SAR descriptors

Abstract

Land cover change detection has long been a hot field in polarimetric synthetic aperture radar (SAR) applications. In certain cases, we care not only the changed areas but also from which type to another. This paper presents a supervised urban land cover change types identification method using a series of polarimetric descriptors from SAR observables and polarimetric decomposition. The normalized difference ratio (NDR) operators are generated firstly using the polarimetric descriptors from SAR images acquired in different dates; these operators are then trained according to the selected training sets and used to identify the remaining samples. A modified superpixel segmentation method for polarimetric interferometric SAR (Pol-InSAR) datasets is introduced to improve the accuracy of the experimental result. Radarsat-2 fully polarimetric SAR (PolSAR) images acquired over Suzhou city, China on 9 April 2009 and 15 June 2010 are used in our experiments, the identification accuracies for all changed land cover types are over 80%, which demonstrates the effectiveness and usefulness of the proposed method.