Through a closer analysis of the mixing process which forms the mixed pixels in multispectral/hyperspectral images and
the physical meanings of fractional abundances, the linear mixing model for the satellite remote sensing image is rebuilt
firstly. Then based on the linear mixing model, a new sub-pixel unmixing method is proposed to estimate the abundance
maps and the spectrums of endmembers. Some experimental results using both simulated hyperspectral data and the real
hyperspectral data show that the new method proposed in this paper is feasible and valid. The estimated abundance maps
and spectrums of endmembers are comparable with that of the real ones.