Classification of Hyperspectral Images by Enhancing Absorption Bands in Spectral Dimension

The spectral signatures in most hyperspectral classification approaches are generally treated as random vectors, which is inappropriate in denoting their typical physical characteristics, such as central wavelengths, widths, and depths of absorption bands. In this paper, we present a new classification approach by enhancing the absorption bands of spectral signatures to boost their physical information. Firstly, an analysis is made of the characteristics of absorption bands of spectral signatures. Next, an absorption bands enhancing approach is proposed based on the discussion of the approach of fusing spectral signatures and their derivative. Finally, the proposed approach is applied on two real hyperspectral subimages. The experimental results show that our proposed approach can significantly enhance the differences of spectral signatures of a hyperspectral images. And thus can improve the classification performance of hyperspectral images.