The increasingly complicated electromagnetic environments make the quality of radar image much poor, and hence impact the downstream task. It is therefore important to suppress the radar interference during imaging. The suppression methods in the echo domain required the details on interference parameters, while those in the image domain were specific to several jamming modes. They were not effective in the practical scenarios. To solve the problem, a novel interference suppression method for SAR images was proposed in this paper. Different from the preceding works, where the explicit modeling of interference signal was built, a new deep learnable architecture with the integration of Swin-Transformer and convolution modules was presented. A simple convolutional network was used to extract the fundamental features from the degraded image. The resulting features were then fed into a pair of networks combine Swin-Transformer and convolutional modules for suppression. The interference components and the suppression results can be obtained accordingly. A loss function composed of the image and the interference measurements was defined to optimize the proposed architecture. To verify the proposed method, several rounds of experiments were performed. The results prove that the imaging quality was improve significantly by proposed method.