Abstract: A deep learning-based phase filtering method is proposed to address phase filtering in interferograms for the application of interferometric synthetic aperture radar technology. A dual-stream network, consisting of a phase filtering stream and fringe detection stream, refered to as PFSFDS Net, is constructed to suppress phase noise in interferograms. Firstly, the phase filtering stream adopts an improved U-Net architecture composed of an encoder and decoder, in which an atrous spatial pyramid pooling (ASPP) module is introduced at the end of the decoder to fuse the fringe information from the fringe detection stream. Secondly, the fringe detection stream is mainly composed of residual blocks that detect fringe information in interferograms and merge it with the phase filtering stream to improve phase filtering for interferograms. Finally, a suitable data set is constructed to enable the trained network to effectively suppress the noise in interferograms while preserving phase information. The experimental results of phase filtering for different types of interferograms show the effectiveness and robustness of this method.