Phase information extraction of moiré fringes based on transfer learning

Moiré fringes contain critical phase information whose precise extraction is of paramount importance. In this paper, a method for moiré fringe phase information extraction based on transfer learning (TL-MFPE) was proposed, with the main purpose of improving training efficiency. The phase unwrapping process is pre-trained, after which the pre-trained model is incorporated into the phase extraction process. Compared with two traditional methods: (1) Fourier analysis and multigrid method (FA + MG), (2) Fourier analysis and quality-guided method (FA + QG), as well as end-to-end deep learning method, our proposed method achieves significantly superior performances in terms of accuracy and noise resistance for moiré fringe phase extraction. Furthermore, constraining the training process with the phase unwrapping procedure compensates for the limited sample size in the dataset, enhancing the model's ability to extract phase information, it is crucial in the training process of real moiré fringes. In a word, this approach is expected to improve the reliability and efficiency of moiré deflectometry, providing more robust technical support for its application.