Abundances Correction in Hyperspectral Data Unmixing to Handle Spectral Variablity

Spectral variability is one of the main causes of reducing hyperspectral unmixing accuracy. Atmospheric condition variability and illumination intensity variability are two main reasons for spectral variability of hyperspectral data. A novel hyperspectral unmixing method based on endmember bundles under sparse regression based unmixing frame is proposed in this paper to handle it. Firstly, the hyperspectral data is corrected to reduce the spectral inconsistency between endmembers from library and imagery caused by varied atmospheric condition. Secondly, the hyperspectral data is unmixed under the sparse regression unmixing frame. Then representation model of spectral variability due to varied illumination intensity is established. Endmember bundles are also established according to this model. Finally, the unmixed fractional abundances are corrected by the endmember bundles. The method in this paper effectively reduces the influences of spectral variability caused by varied atmospheric condition and illumination intensity and promotes unmixing accuracy of hyperspectral data.