A Two-Dimensional InSAR Phase Unwrapping Method Based on Deep Convolutional Neural Network and L1 Norm Optimization

Abstract

Synthetic Aperture Radar Interferometry (InSAR) is a technology that obtains three-dimensional information on the surface of the earth by calculating the phase difference of the interferometric complex image. Phase unwrapping plays a decisive role in the accuracy of digital elevation model. Estimating the true phase gradient is a key step in phase unwrapping. The traditional phase unwrapping algorithm depends on the continuity assumption to calculate the unwrapping phase gradient. However, due to the discontinuity caused by noise and terrain mutation, the unwrapping phase will have a large error. In this paper, a phase unwrapping method based on deep convolution neural network and L1 norm optimization is proposed. Firstly, the phase ambiguity gradient is calculated by feature pyramid and global attention network (FGANet), which is a lightweight network combining two attention mechanisms. Then, the neural network calculation results are corrected by L1 norm to obtain accurate phase ambiguity gradient data. Finally, the unwrapping phase data can be obtained by simple integration. By using the FGANet network to calculate the phase ambiguity gradient instead of the original phase continuity assumption, it has higher accuracy and robustness than traditional algorithms. Can quickly unwrap a wide range of wrapping phases.