Correction of spurious phase sign in single closed-fringe demodulation using transformer based Swin-ResUnet.

• The phase detection of single frame interference fringe provides the possibility of dynamic phase measurement. • Traditional image segmentation methods can not effectively segment the inverted region of phase sign. • We first applied the deep learning based on Swin Transformer structure to correct the spurious phase sign inversion in single closed-fringe demodulation. • We compared with the classic Convolutional neural network, the network based on Swin Transorfermer structure has better generalization ability. Single-shot fringe analysis draws extensive attention dues to the possibility for dynamic phase measurement. However, phase extraction from single interferogram with closed fringes is still a difficult task, the dominate reason is spurious phase sign problem. The fringe direction map is usually required to unambiguous the spurious phase sign; however, it is not a trivial process even though numerous algorithms are able to estimate the fringe direction. Swin Transformer has shown great promise as it integrates the advantages of both CNN and Transformer. In this contribute therefore, we propose to unambiguous the spurious phase sign problem using transformer based Swin-ResUnet which has strong feature extraction capabilities. Proposed method is insensitive to noise and easy to perform, the simulation and experiment corroborate its effectiveness.© 2023 Elsevier Ltd. All rights reserved. [ABSTRACT FROM AUTHOR]