Abstract: In order to fully exploit the intrinsic properties of abundance maps and improve the sparse unmixing accuracy of hyperspectral image, a sparse unmixing method based on local weighted low rank prior is proposed. The low rank prior is mainly based on the fact that the local cubes in hyperspectral images have higher spatial correlation and spectral correlation. The weighted low rank prior can explore the low-dimensional structural features inherent in the local block, effectively suppress the noise and maintain the detailed structure of the data. In combination with the existing total variation regularization and collaborative sparse regularization, the proposed method shows improved ability to exploit the detailed structure of the abundance coefficients, local smoothness and row sparsity. Experimental results on the simulation data and real hyperspectral data show that the proposed method can better maintain the fine details of the data and improve the unmixing accuracy compared with other state-of-the-art methods.

Key words: sparse unmixing, total variation, weighted low rank, collaborative sparse, alternating directions method of multipliers (ADMM)