Research On The Application Of Multi-scale Feature Fusion Residual Network In InSAR Phase Unwrapping

Synthetic Aperture Radar Interferometry(InSAR)is a technology that integrates interferometry and SAR imaging.As a crucial link in the InSAR technical process,phase unwrapping can restore the real phase by wrapping the phase,thereby obtaining the terrain inversion result.On the one hand,the application of deep learning technology is profoundly affecting technological changes in various fields;on the other hand,there are some urgent problems to be solved in phase unwrapping technology,which requires some high-precision and high-efficiency unwrapping algorithms.Phase unwrapping research has become a trend.One is based on the improvement of the branch-cut method.The shortest branch-tangent line between the residual points is obtained through the ant colony algorithm,so as to solve the "island" phenomenon of disentanglement in the branch-cut method.The experimental results show that the length of the branch-cut line is shortened by more than 50%,but Less improvement in unwrapping efficiency.In order to better extract image features,firstly,a multi-scale feature extraction module is added on the basis of the residual network,and the structure of the residual module is improved to improve the generalization ability of the network;secondly,the CBAM attention mechanism module and the ResDW module are introduced to Obtain more semantic information and enhance feature extraction ability;finally,the construction method of InSAR data set is studied,and an interferogram data set composed of simulated terrain data,measured terrain data and DEM data is created.Training and testing are performed on the created dataset to demonstrate the effectiveness of the method.Experiments are carried out on the simulated terrain data and the measured terrain data.The root mean square error and the unwrapping time are used as the evaluation indicators.The experimental results show that there is a significant improvement in the unwrapping efficiency and the unwrapping accuracy.This paper has 37 figures,6 tables and 57 references.