Abstract:

In order to make full use of the feature information contained in the reconstruction of the sparse representation classification algorithm, the band information of the reconstructed residual is fed back into the test sample to enhance the extraction of the feature information. However the feature may be over fitted in the feedback adjustment process. In order to further improve the stability and the classification accuracy of the proposed algorithm, the close coupled set of pixels (CCSP) generation algorithm is proposed to avoid the over fitting by smoothing the distribution of the feature. Finally, the close coupled set of pixels-based adaptive boosting class-wise sparse representation classifier (CCSP-ABCWSRC) algorithm is proposed. Experimental results based on Indian Pines, 〖JP2〗University of Pavia, Salinas three hyperspectral data sets show that the proposed algorithm is effective for hyperspectral images classification and its classification accuracy is better than the similar algorithm.