A hybrid sparsity and constrained energy minimization detector for hyperspectral images

摘要

Sparse representation has been successfully used to solve target detection problem in hyperspectral images (HSI). Compared with the traditional target detection methods, it is not fully dependent on statistical structure of the data sets. In this paper, a hybrid sparsity and constrained energy minimization (HSCEM) detector for HSI is proposed. In sparse representation, local clustering or unmixing is used to obtain the dictionary, and the greedy subspace pursuit (SP) algorithm is used for sparse representation coefficient estimation. Combining sparsity-based detector with the traditional statistics-based detection method (CEM detector), the reconstructed result rather than reconstruction error is employed to distinguish between target and background. Experimental results illustrate the outperformance of the proposed HSCEM detector over several classic statistics-based detectors and sparsity-based detectors.