Enhanced single-frame interferogram phase retrieval using a model-based domain adaptation network

Abstract

Accurate phase retrieval from interferograms is critical for interferometry. Existing deep learning methods fail to fully exploit the physical model of the interferometry, resulting in limited accuracy. This paper proposes a model-based domain adaptation network (MDANet) for end-to-end phase retrieval from single-frame interferograms. MDANet effectively extracts domain-invariant phase features, enhancing its adaptability and robustness across diverse interferometric systems. The dataset generated using the model-based approach facilitates enhanced learning of phase features. The network architecture consists of an encoder, a discriminator, and a decoder. The encoder extracts phase features from the interferograms, while the discriminator reduces the domain gap, ensuring that only phase information is preserved. The decoder reconstructs these features into output phase maps. Simulation and experimental results demonstrate that MDANet outperforms existing methods in accuracy and adaptability, offering an improved solution for dynamic interferometry.