Abstract: Most of the single image rain removal methods based on deep learning regard rain removal as a simple end- to-end mapping problem, and do not make full use of the inherent prior information of the image, resulting in unsatisfactory rain removal effect. In this regard, a single image rain removal network is proposed that combines image frequency priors and conditional generative adversarial networks. The generator of this network was designed based on improved U-Net, and the frequency information of the image was introduced into the discriminator as a conditional constraint to guide the generator to generate clear rain removal results. The experimental results show that the qualitative and quantitative evaluation results of this method are better than the current mainstream methods.