Deep learning phase-unwrapping method based on adaptive noise evaluation

Abstract

To address the problem of phase unwrapping for interferograms, a deep learning (DL) phase-unwrapping method based on adaptive noise evaluation is proposed to retrieve the unwrapped phase from the wrapped phase. First, this method uses a UNet3+ as the skeleton and combines with a residual neural network to build a network model suitable for unwrapping wrapped fringe patterns. Second, an adaptive noise level evaluation system for interferograms is designed to estimate the noise level of the interferograms by integrating phase quality maps and phase residues of the interferograms. Then, multiple training datasets with different noise levels are used to train the DL network to achieve the trained networks suitable for unwrapping interferograms with different noise levels. Finally, the interferograms are unwrapped by the trained networks with the same noise levels as the interferograms to be unwrapped. The results with simulated and experimental interferograms demonstrate that the proposed networks can obtain the popular unwrapped phase from the wrapped phase with different noise levels and show good robustness in the experiments of phase unwrapping for different types of fringe patterns.