Adapted sparse fusion with constrained clustering for semisupervised change detection in remotely sensed images

Abstract

Change detection is defined as the process that helps in determining the changes associated with land use and land cover properties with reference to georegistered multitemporal remote sensing data. A unique idea of integrating sparse fusion with constrained clustering has been attempted to detect changes in multitemporal and multispectral remote sensing images in a framework called ASCC (to be pronounced as “ask”). Unlike the other traditional methods of change detection, this approach focuses on the fusion of difference images followed by a semisupervised clustering to construct a change detection map in binary form. An enhanced sparse representation is adopted to improve the performance of the fusion process by creating a locally adaptive dictionary that extracts patches from the two difference images. The reconstruction of the fused image is performed using maximum absolute coefficients of the learned dictionary. The resultant fused image is subjected to the change detection process through constrained clustering to discern the changed pixels from the unchanged pixels. To evaluate the designed solution, the method is quantified through percentage correct classification and the process is compared with the existing methods of clustering such as k-means, adaptive k-means, fuzzy C-means, and enhanced constrained k-means. The simulation results demonstrated that the designed approach performed better when compared with state-of-art change detection methods for detecting changes in multitemporal images.