According to the seasonal variation of wetland and the difference of vegetation cover in different wetland types, the multi-temporal GF1-WFV remote sensing data and polarization features of GF3-FSⅡ data were comprehensively used to research on wetland fine classification method. Firstly, totally 50 high importance feature values from the spectral information, vegetation index and water index of thirteen GF1-WFV remote sensing data were optimized, which used OOB sample of random forests, and the 50 high importance feature values were used to preliminary classification of wetland. Then, aiming at the problem that marsh grassland, shrub swamp and marsh land were mixed in the classification results, and the recognition accuracy of some wetland types was low, the separation degree of backscattering features was analyzed from two dimensions of intensity and amplitude by using the GF3-FSⅡ HH and HV polarization data of vigorous vegetation growth, optimized σFD-HH feature was used for wetland types classification. Finally, taking Da’an City in Jilin Province as the research area to verify and analyze the method. The results showed that the overall accuracy of wetland classification reached 86.23% with Kappa coefficient of 0.82. The results can provide technical support for wetland resource investigation and management.