Feature Pyramid and Global Attention Network Approach to Insar Two-Dimensional Phase Unwrapping

Abstract

Phase unwrapping is a key step of InSAR signal processing. The traditional algorithm relies on the phase continuity assumption. However, the unwrapped phase is not usually effective because of severe noise and dense fringes. A phase unwrapping method based on deep convolution neural network is proposed. The phase ambiguity gradient is calculated by Feature Pyramid and Global Attention Network (FPGANet), which is a lightweight network combining two attention mechanisms. The neural network calculation results are utilized by a postprocessing to obtain the final unwrapped phase. The first test is conducted on simulated data, which is generated from a different signal model. The second test shows the new method is time efficient as the interferogram size increases. The experimental results indicate the effectiveness of FPGANet.