Contextual and Spectral Feature Fusion Using Local Binary Graph for Hyperspectral Images Classification

Abstract

There are several approaches for hyperspectral (HS) image classification. But the best approach is using contextual and spectral information concurrently. A major challenge for researchers is fusing spectral and spatial contextual features. LBG (Local Binary Graph) is an efficient technique among them. An enhanced type of LBG is suggested in this paper, which involves the class label for feature extraction to minimize within class similarity. The proposed method considers three constraints for selection of the nearest spectral-spatial contextual neighbors and sharing between them. The constraints include the minimum distance of the spectral features vector, minimum distance of the spatial contextual features vector and the belongings to the same class. So, the proposed method can fuse the spectral and spatial contextual features with increasing the class discrimination ability. The experiments indicate that the proposed method improves the overall classification performance on Pavia University and Indian pines data sets.