Structured sparse Bayesian hyperspectral compressive sensing using spectral unmixing

To reduce huge consumption of processing hyperspectral images(HSI), a novel Bayesian unmixing compressive sensing framework is proposed to compress and reconstruct HSI effectively, called structured sparse Bayesian umixing compressive sensing(SSBUCS). SSBUCS unites compressive sensing and hyperspectral linear mixed model in Bayesian framework. An HSI is decomposed as a linear combination of endmembers and abundance matrix. The abundance matrix is transformed to a structured sparse signal in the wavelet domain. Then, compressive sensing is employed on this sparse signal to produce a more compact result. To recover the HSI, a Markov chain Monte Carlo(MCMC) method based on Gibbs sampling is proposed, imposing structured sparse prior on abundance matrix. Experimental results verify the superiority of the proposed method over several state-of-art methods.