Multiple Endmember Extraction Using Spatial-Spectral Information

Spectral variability often occurs in hyperspectral remote sensing scenes. The traditional hyperspectral endmember extraction method ignores the influence of spectral variability, resulting in low accuracy of mixed pixel decomposition. In this work, a multiple endmember extraction algorithm based on spatial-spectral information is proposed for hyperspectral image(HSI). Firstly, the original hyperspectral image is divided into non-overlapping image blocks, and the traditional single endmember extraction algorithm is used to extract candidate endmembers on the image blocks. Secondly, the spectral similarity analysis is performed on the neighborhood pixels of the candidate endmembers to refine candidate endmembers. Finally, the clustering algorithm is used to cluster all endmember sets to obtain the final multiple endmember set. The experimental results show that the proposed algorithm has the ability to characterize the spectral variability in remote sensing images and can effectively improve the accuracy of mixed pixel decomposition.