Hyperspectral unmixing based on iterative weighted L1 regularization

In order to improve the accuracy of hyperspectral unmixing, linear unmixing based on sparsity is studied. A novel method of linear hyperspectral unmixing based on iterative weighted L1 regularization is proposed, and the corresponding model and algorithm are presented. The method introduces several steps of weighted L1 optimization procedures, and uses the value of current solution to revise the weights for next iteration, which makes the sparsity of fractional abundances of mixed pixel be represented better. Experimental results demonstrate that the accuracy of hyperspectral unmixing based on iterative weighted L1 is higher than traditional L1 regularization, especially for high signal-to-noise ratio hyperspectral images.