Adaptively group based on the first joint sparsity models distributed compressive sensing of hyperspectral image

Hyperspectral Images (HSI) have strong spectral correlation compared with ordinary 2D images. Distributed compressed sensing (DCS) happens to exploit both intra-and inter-signal correlation structures among multiple nodes and lends itself well to hyperspectral image compression. In this paper, we propose a new algorithm of adaptive grouping for HSI compression based on the first joint sparsity model (JSM-1) of DCS. This algorithm adaptively divides one hyperspectral image into several groups of bands (GOBs) in accordance with its spectral correlation firstly, to ensure that each group of bands has strong spectral correlation. Every group of bands contains a reference band and the remaining non-reference bands, and then subtracts the reference band from each of the non-reference bands in the same group which makes the structure conform JSM-1. Then the distributed compressed sensing JSM-1 model is applied to hyperspectral image compression, encoding every residual image using CS coding. We use a joint recovery algorithm to reconstruct at the decoder. In this algorithm, the spectral similarity of high spectral images is used to get the data more sparse and improve the reconstruction effect of the compressed image, and the better compression efficiency is obtained. Experiments show the feasibility of the proposed algorithm.