Abstract: The hyper-spectral features (red edge, green peak, red valley) of typical desert grasslands in Xinjiang Fukang City were extracted from hyper-spectral lines through differential and Savitzky-Golay smoothing filter processing. The desert vegetations have absorption features, such as red edges and green peaks but the hyper-spectral curves are not clear because of the effect of the underlying surface showing through the sparse foliage of the low coverage desert vegetation; Due to the withered litter desert vegetation in October, the spectral reflectance effect of soil interference was too large, and compared with that in May, the red edge position showed the phenomenon of “red edge shift”. The NDVI (the normalized difference vegetation index) and RVI (ratio vegetation index) index values were lower in October. There were significant differences in red edge, green peak, red peak and vegetation index characteristics between grassland types, different seasonal desert vegetations, but NDVI and RVI were associated with strong correlations. Finally, the feature extraction and analysis of hyper-spectral data from typical desert grasslands in Xinjiang not only enriches the means of hyper-spectral data mining, but also provides a scientific basis for grassland classification based on hyper-spectral data.