Sparse representation-based hyperspectral image classification

Hyperspectral image (HSI) contains diagnostic continuous spectrum, which benefits the precision land-cover classification. However, there is also high correlation between adjacent bands, and redundancy in both the spectral and spatial domains. Therefore, it has been demonstrated that HSI is essentially low-rank and can be represented sparsely. This chapter firstly reviewed the classic representation-based models, including the sparse representation-based classifier (SRC) and collaborative representation classifier (CRC). Then, a series of original work on SR-based framework are carried out from two aspects. One is the improvement and exploration in the spectral domain, with respect to the features extraction and decision mechanism. The other one is the collaboration and integration in the spatial-spectral domain, with respect to the utilization of spatial information in the spatial-spectral domain. The experimental results based on two real hyperspectral data sets demonstrate their efficiency, with improvements over the other related methods.