基于谱间结构相似先验的高光谱压缩感知重构

Hyperspectral Compressive Sensing Recovery via Spectrum Structure Similarity

在高光谱压缩感知重构中,充分利用图像的先验信息能有效提升算法的重构精度。现有重构算法均未考虑高光谱图像的谱间结构冗余信息,该文提出一种基于谱间结构相似先验的高光谱压缩感知重构方法。该方法通过谱间结构冗余定义高光谱结构图像,以结构图像为基础,设计一个压缩感知重构正则项,再结合高光谱图像的空间相关性和谱间统计相关性,提出一种新的压缩感知高光谱图像联合重构方案,并设计一种基于变量拆分的有效的求解算法。实验表明,在相同观测值数目下,该文算法的重构质量明显优于现有算法。

In the hyperspectral compressive sensing reconstruction method, the exploitation of the prior information of the hyperspectral imagery can improve the reconstruction performance. As the existing methods have not taken into account the spectral structural redundancy information of hyperspectral imagery, a novel reconstruction method via spectrum structure similarity for hyperspectral compressive sensing is proposed in this paper. Structure images are proposed via spectrum structure similarity and a new regularizer is given based on structure images. It combines the new regularizer and other regularizers,so that the spatial redundancy, spectral statistical redundancy and spectral structural redundancy in hyperspectral imagery can all be exploited. In addition, an efficient solving algorithm based on variable-splitting is developed for the method. Experimental results show that the proposed method is able to reconstruct the hyperspectral imagery more efficiently than the current methods at the same measurement rates.