Abstract:

To take full advantage of the information of the sparse representation classification algorithm and the spatial information of the hyperspectral image, a weighted conditional sparse representation combined with the Markov random field (MRF) algorithm is proposed. Firstly, a conditional sparse representation model is built for the residual vectors, which introduces the frequency band variance information into the problem of calculating the reconstructed error. Then the spectral information divergence is used to take advantage of the information of the reconstructed spectrum from the perspective of information entropy. The conditional sparse representation model is weighted by the spectral information divergence in the expectation maximization algorithm, which makes the proposed algorithm have the ability of self renew. On the condition of the same convergence rate, the MRF is introduced into the weighted conditional sparse representation algorithm to utilize the spatial information of the hyperspectral image. Simulation results show that the propose method can effectively improve the classification accuracy, and has good stability in different experimental data.