Blind separation of component information from mixed pixels in hyperspectral imagery

Mixed pixel separation has always been a hotspot issue of quantitative remote sensing and is especially important for hyperspectral imagery. A new method based on independent component analysis is proposed in this paper to blind separate component information including the signature and the weight from the spectrum of mixed pixels. The extra information is obtained from the statistical characters of the signatures. To evaluate the performance of the algorithm some computer numerical simulations are conducted and a method to choose a best band coverage of the spectrum using for blind signal separation is proposed. Finally the algorithm is applied on the HYPERION imagery of Henshan, Shanxi Province and the result showed that the method is effective.