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

Area of Science:

• Optics and Photonics
• Metrology
• Artificial Intelligence in Imaging

Background:

• Electronic Speckle Pattern Interferometry (ESPI) is a powerful non-contact measurement technique.
• Phase unwrapping is critical for ESPI precision but is often degraded by speckle noise.
• Conventional algorithms struggle with phase continuity under high noise levels.

Purpose of the Study:

• To develop a noise-robust deep learning architecture for accurate phase recovery in ESPI.
• To enhance the reliability of ESPI measurements in the presence of significant speckle noise.
• To reduce the domain gap between synthetic and real-world ESPI data.

Main Methods:

• Proposed U-ResNet-ESPI, a deep learning architecture integrating physics-driven data synthesis.
• Implemented a physics-aware multi-view preprocessing step.
• Utilized an optimized U-Net with a topology-aware hybrid loss function.

Main Results:

• Achieved highly accurate phase recovery even under severe speckle noise conditions.
• Demonstrated a low error level of only 0.01 radians.
• Significantly outperformed traditional phase unwrapping algorithms in robustness and accuracy.

Conclusions:

• U-ResNet-ESPI offers a robust solution for accurate phase recovery in ESPI.
• The method enhances the reliability of ESPI measurements, particularly in noisy environments.
• The approach is extensible to other coherent imaging techniques like digital holography and stereography.