Decomposition of Mixed Pixels Based on Fisher Discriminant Null Space in Hyperspectral Imagery
基于Fisher判别零空间的高光谱图像混合像元分解

In the traditional spectral mixture analysis method, each endmember is assumed to have a constant spectral signature. However, there are always differences in endmember spectra for same ground objects. In order to restrain the effect of different endmember spectra of the same object on mixed image decomposition, a new method based on Fisher discriminant null space for decomposition of mixed pixels in hyperspectral images is proposed. In the Fisher discriminant null space method, the linear transformation of hyperspectral image data can make the endmember spectra have no variability inside each endmember group and have large differences among different endmember groups. Therefore, the negative impact resulted from the endmember spectral variability can be decreased to a larger extent by using the transformed spectra. The experimental results of both artificial data and actual remote sensing images in Indiana and Cuprite regions show that the proposed algorithm has a higher decomposition accuracy for the mixed pixels in hyperspectral images.