A Synthetic Aperture Radar (SAR) image target classification method is proposed by combining Bi-dimensional Variational Mode Decomposition (BVMD) with Convolutional Neural Networks (CNN).The original SAR images are decomposed by BVMD into a series of modes,which reflects the global and detail information of the targets.Afterwards,a suitable CNN architecture is designed to classify different modes and output the posterior probability vectors.Then,those posterior probability vectors are fused by using the Bayesian theory.Finally,the target label is decided according to the fused probabilities.Through the combination of the advantages of BVMD and CNN,the proposed method improves the classification performance comprehensively.In the experiments,the proposed method is tested based on MSTAR dataset under four typical scenarios and compared with the existing methods.The results show the superiority of the proposed method.