Semi-supervised change detection approach combining sparse fusion and constrained k means for multi-temporal remote sensing images

Change detection is the measure of the thematic change information that can guide to more tangible insights into an underlying process involving land cover, land usage and environmental changes. This paper deals with a semi-supervised change detection approach combining sparse fusion and constrained k means clustering on multi-temporal remote sensing images taken at different timings T1 and T2. Initially a remote sensing fusion method with sparse representation over learned dictionaries is applied to the difference images. The dictionaries are learned from the difference images adaptively. The fused image is calculated by combining the sparse coefficients and the dictionary. Finally the fused image is subjected to constrained k means (CKM) clustering combining few known labelled patterns and unlabelled patterns which have been collected from experts. The enhanced (CKM) approach (ECKM) is compared with k means, adaptive k means (AKM) and fuzzy c means (FCM). Experimental results were carried out on multi-temporal remote sensing images. Results obtained using PCC and F1 measure confirms the effectiveness of the proposed approach. It is also noticed that the ECKM provides better results with less misclassification of errors as compared to k means, adaptive k means and fuzzy c means.