Hyperspectral Unmixing Algorithm Guided by Multiscale Spectral Similarity

Sparse Unmixing (SU) combined with spatial information constraints is one of the research hotspots in the field of Hyperspectral Unmixing (HU). In order to overcome the shortcomings that the spatial structure of hyperspectral images in natural scenes is difficult to accurately represent. Hence, a Hyperspectral Unmixing algorithm guided by multiscale spectral similarity was proposed in this paper. First, segment the Hyperspectral Image into an approximate domain spectral image with a spatial structure. Then, the approximate domain SU was performed according to the similarity between adjacent superpixels. Finally, the approximate domain unmixing result was converted to the original domain, and combined with the actual pixel spectrum to perform accurate pixel-by-pixel unmixing of the original domain. Experimental results show that the algorithm proposed has lower unmixing complexity than similar algorithms, and has higher unmixing accuracy and robustness.