Application of super-pixel segmentation in sparse unmixing of hyperspectral image

• 1、College of Information and Communication Engineering, Harbin Engineering University, Harbin 154000

Abstract：The classical collaborative sparse unmixing algorithm considers that all pixels in a hyperspectral image share the same set of endmembers, adding a global collaborative sparse constraint to the abundance matrix of the solution. However, since the endmembers in the hyperspectral image are always presented in the spatially homogeneous region rather than the entire image scene, only the abundance corresponding to the pixels in the local homogeneous region has synergistic sparsity. In order to more effectively contain local spatial information in the sparse unmixing model, this paper proposes a super-pixel-based local collaborative sparse unmixing (SP-LCSU) algorithm, mainly compares unmixing performance of the local collaborative sparse unmixing algorithm based on quaternion color distance super-pixel segmentation algorithm (QSP-LCSU) and the local collaborative sparse unmixing algorithm based on simple linear iterative clustering super-pixel segmentation algorithm (SSP-LCSU). Compared with the existing sparse unmixing algorithm, the synthetic data and real hyperspectral data experimental simulation verify the effectiveness and superiority of the proposed algorithm.

Keywords： signal and information processing； hyperspectral image； sparse unmixing； local spatial information； superpixel segmentation