Due to the similarity of backscattering intensity of rocks

the general SAR data is not easy to make lithological classification on the earth surface. Polarimetric Radar records Stokes of Scattering Matrix of backscattering waves

which increases the classification precision. But meanwhile

the data fusion coming from the coherence of different polarizations in the polarimetric data produces more error for lithological classification. This paper divides the rock’s backscattering into three parts: Single Scattering

Double Scattering

and Cross Scattering The decomposition decreases the coherence of co polarization backscattering waves

enhances the backscattering difference of rocks

and shows the ability of lithological classification for polarimetric data. The polarimetric data used in this paper was obtained in October 1993

in the north of Xinjiang Province. With the statistic analysis

this paper builds the scatter plots for each kind of rocks

and input them into the Neural Network Classifier. The maximum precision is 94% and minimum error 1.06%.