unsupervised change detection

Over the last two centuries, natural hazards and widespread human activities have led to significant changes in water, soil, and vegetation covers. Although, multi-temporal remote sensing (RS) images provide continuous monitoring of changes in land surface, one of the most important challenges is applying multi-sensor images to detect land cover changes in unsupervised flow. This study aims to provide a method for changes detection in water, soil, and vegetation covers within multi-sensor RS images. In this regard, new biophysical parameters have been defined for the Sentinel2B sensor, as well as a new unsupervised method for binary and multiple changes detection has been developed. Landsat8 OLI and Sentinel2B images of the southwestern shore of the Urmia Lake were used to evaluate this method. In the proposed method; first, the generalizability of the Tasseled Cap (TC) transformation was investigated and a new TC transformation for the Sentinel2B sensor was estimated. After TC, the images were transferred from multispectral feature space to biophysical feature space, and a binary changes map was generated using the proposed multivariate iterative trimming method. Then, via FCM clustering, the changed samples were separated into a certain number of clusters determined by the WSJI criterion which is one of the innovations of the proposed method. Overall accuracy, missed error, and false alarm of the proposed approach are %92.06, %9.62, and %6.27, respectively. The proposed method in this paper can be used as an unsupervised, accurate, and reliable technique for changes detection in water, soil, and vegetation covers.