Hyper-spectrum classification based on sparse representation model and auto-regressive model

A novel classification approach based on sparse representation model and auto-regressive model is presented to deal with spectral and spatial information underutilization effectively for hyper-spectrum classification. The combination dictionary is designed using sparse representation model and auto-regressive model. Sparse representation model is used to represent every spectral vector as sparse linear combination of the training samples on spectral dimension; auto-regressive model is added to constrain every spectral vector by its eight neighborhoods on spatial dimension. A new dictionary is constructed for every class to reduce the computation and reconstruction error. At last, the sparse problem is recovered by solving a constrained optimization of minimum reconstruction error and neighboring relativity. The classification of hyper-spectral image is determined by computing the minimum reconstruction error of testing samples and training samples. Simulation results show that the method improves the classification accuracy.