ABSTRACT
Spectral variability existing in hyperspectral (HS) images reduces the accuracy of unmixing. To mitigate the effect of spectral variability on unmixing, we propose a spatial-spectral clustering-based method (called SP2C) for endmember extraction and HS unmixing. The proposed SP2C adopts a local spatial-spectral clustering strategy to obtain a set of spatially homogenous regions and exploits a spectral purity index calculation strategy to choose pure pixels on the regions, where the average spectra of chosen pure pixels on the regions are taken as endmember candidates to alleviate local spectral variability. Then, an adjusted k-means plus (called AD k-means++) algorithm is employed to cluster candidate endmembers to alleviate global spectral variability, and the final endmembers are achieved by matching the cluster centers and the reference endmembers obtained by the vertex component analysis (VCA)-like methods. Our experiment results, conducted using real HS datasets, confirm that the proposed method considerably improves the HS unmixing performance compared to the state-of-the-art techniques.
Disclosure statement
The authors declare no conflict of interest.