Hyperspectral remote sensing is increasingly used to determine the feature components of mixed pixels and their proportions. In this paper, the vegetation soil mixed pixels of different area ratio were set as the research object, and polarization means and ASD FieldSpec3 spectrometer was applied to obtain polarized reflectance spectrum curve of vegetation soil mixed pixels to calculated the proportion of 8 different vegetation index and discuss the hyperspectral polarization characteristics of vegetation soil mixed pixels under different area ratio and different polarization angle. The study found that as the increasing of the proportion of leaves, vegetation soil spectral curves increasingly appeared “5 valleys and 4 peaks”, and the positions of the peak and bottom were basically the same. The larger the angle of polarization, the greater the spectral reflectance ratio of mixed pixels was; In mixed pixels, the larger the proportion of the vegetation, the greater the influence of the polarization angle. The vegetation index and the size of vegetation in mixed pixels were in a linear relationship and the correlation coefficient between the vegetation attenuation index and the improved red edge normalized difference vegetation index was the largest, which could reach about 98%, suitable for establishing the correlation model between vegetation index and vegetation proportion of mixed pixel area.. When vegetation area changes, the sensitivity of vegetation index is better by improving the red edge ratio. In the use of spectral absorption characteristic parameters to estimate vegetation index, the two order function model of absorption valley depth and photochemical vegetation index had the strongest fitting degree with the determination coefficient R2 of 0.963 3; The two degree function model of spectral absorption index and photochemical vegetation index had the strongest fitting degree, and the coefficient of determination R2 was 0.960 5.