Phase unwrapping with the convolutional neural network

Abstract

Phase unwrapping is a classical signal processing problem, which refers to the recovery of the original phase value from the wrapped phase. Two dimensional phase unwrapping is widely used in optical measurement technology, such as digital holographic interferometry, fringe projection profilometry, synthetic aperture radar and many other applications. In this paper, a phase unwrapping method with the convolution neural network is proposed, and the feasibility is analyzed by numerical simulation. The convolution neural networks with different parameters are set up, and the phase screens used for the training set and testing set of convolution neural network are simulated with MATLAB software. The numerical simulation results show that the four convolution neural network models can be used for phase unwrapping, but the parameters have a significant impact on its accuracy.