A weighted multi-task joint sparse representation method for hyperspectral image classification

In this paper, a novel weighted multi-task joint sparse representation method is proposed for hyperspectral image classification. It is assumed that the importance of atoms in a dictionary can be weighted when they are used in sparse representation according to the similarities between tasks and classes. We utilize tasks instead of classes in pre-classification to group all samples into several clusters, one cluster stands for one tasks. Then we calculate the correlations between tasks and labels based on the reconstruction errors of sparse representation acquired from training samples. The correlations are used for the consideration of heterogeneous neighborhood, which is the core of multi-task method. The weights of different tasks can be adjusted using training samples according to the reconstruction errors. At last, all samples can be classified more accurately via task correlations and reconstruction errors. Experimental results on real hyperspectral data sets exhibit its superiority to compared algorithm.