Abstract: Interferometric phase filtering is one of the key steps in Interferometric Synthetic Aperture Radar (InSAR) measurement data processing.The accuracy of the results will directly affect the accuracy of phase un-wrapping,and then affect the final InSAR data processing accuracy.In recent years,deep learning InSAR inter-ferometric phase filtering has received ex tensive attention due to its superior performance,but its model still has bottlenecks such as weak generalization ability and poor ability to maintain interferometric phase details.Therefore,an InSAR interferometric phase filtering method based on deep re sidual network is proposed.Based on the original residual network structure,this method integrates the channel attention mod ule to enhance the generalization ability of the deep learning interferometric phase filtering network model and enhance the de tails of the deep learning interferometric phase filtering results.Aiming at the problem of edge phase jump of interference frin ges,the network model takes the real and imaginary parts of the deformed interference phase as input,so as to effectively main tain the edge information of dense fringes.The simulation data and the measured data are used for experiments,and compared with the existing spatial domain,frequency domain and deep learning filtering methods.The experimental results show that the proposed algorithm improves the root mean square error (RMSE) by nearly 25% and the structural similarity index (SSIM) by nearly 10% in the simulation data.In the measured data,the proposed method can better retain the phase detail information.

Key words: interferometric phase filtering,deep learning,residual network,channel attettion