Fusion of SAR and Optical Data for Unsupervised Change Detection: A Case Study in Beijing

Change detection can either be carried out using multitemporal optical or synthetic aperture radar (SAR) images. Due to the different electromagnetic spectrum used, these two types of imagery provide different representations of the same physical reality. Change information extraction can benefit from the fusion of SAR and optical data. In this paper we investigate the fusion of SAR and optical for change detection application. Beijing, the capital of China that has experienced rapid urbanization, is selected as a case study. Two multitemporal datasets that consist of Landsat and SAR (ERS-2 and ENVISAT) images are used. An unsupervised classification framework that combines the virtues of the k-mean and SVM supervised classifier is proposed. Different fusion strategies are tested including fusion at the feature level and at the decision level. The analysis reveals that the best result can be obtained when the fusion of change information is carried out at the decision level.