Joint segmentation and classification of hyperspectral image using meanshift and sparse representation classifier

A novel spectral-spatial classification method based on mean shift and sparse representation classifier (SRC) for hyperspectral images is proposed in this paper. Firstly, the nonnegative matrix factorization, is used as a preprocessing for mean shift. Then, the mean shift algorithm is adopted to partition an image into amount of blocks and get the segmentation map. Through this way, many size-variable and close regions can be got while the boundary information is remained. Secondly, the classification map is obtained by using the SRC. Finally, the fusion of the segmentation map and the classification map is done by using the majority vote rule. Experimental results on two real hyperspectral images demonstrate the effectiveness and good performance of the proposed method.