Abstract: Aiming at the problem of low feature information and noise interference extracted from polarimetric synthetic aperture radar (POLSAR) images, in this paper a multi-scale fusion method for different feature information is proposed, and the fused features are used to classify target by deep learning network algorithm. Firstly, feature components obtained from polarization decomposition such as Freeman and Yamaguchi are fused on different scales by wavelet transform technology, and then principal component analysis(PCA) dimension reduction algorithm is used to process the fused data, and finally DeepLabV3 network structure is input for training. Moreover, the full-polarization SAR data of Gaofen-3 in baihu farm area is used for verification, and the classification results before and after fusion are compared. The classification accuracy of the algorithm proposed in this paper has been improved significantly, which proved the effectiveness of the method.