U-ResNet-ESPI: a physics-informed deep learning framework for robust phase unwrapping in electronic speckle pattern interferometry

Electronic speckle pattern interferometry (ESPI) is a non-contact, full-field measurement technique whose precision hinges on robust phase unwrapping. However, high-speckle-noise often compromises phase continuity in conventional algorithms. To address this, we propose U-ResNet-ESPI, a noise-robust deep learning architecture for accurate phase recovery. It combines physics-driven data synthesis, a physics-aware multi-view preprocessing, and an optimized U-Net with a topology-aware hybrid loss to reduce the synthetic-to-real domain gap. improve robustness to varying levels of speckle noise, and preserve phase continuity. Experimental results show that the proposed method achieves accurate phase recovery even under severe noise conditions with an error level of only 0.01 rad, significantly outperforming traditional algorithms. This approach improves the reliability of ESPI measurements and can be extended to other coherent imaging modalities such as digital holography and stereography.