Deep learning phase-unwrapping method based on adaptive noise evaluation.

Area of Science:

• Optics and Photonics
• Image Processing
• Artificial Intelligence

Background:

• Phase unwrapping is crucial for interpreting interferometric data.
• Existing methods struggle with noise and varying fringe patterns.
• Accurate phase retrieval is essential for applications like deformation measurement.

Purpose of the Study:

• To develop a deep learning-based phase unwrapping method.
• To incorporate adaptive noise evaluation for improved accuracy.
• To enhance robustness across different noise levels and fringe types.

Main Methods:

• A UNet3+ architecture combined with a residual neural network was employed.
• An adaptive noise level evaluation system integrated phase quality maps and residues.
• Deep learning networks were trained on diverse noisy datasets.

Main Results:

• The proposed method successfully retrieves unwrapped phases from noisy interferograms.
• Demonstrated good robustness with simulated and experimental data.
• Effective for various fringe pattern types and noise levels.

Conclusions:

• The deep learning approach with adaptive noise evaluation offers a robust solution for phase unwrapping.
• This method significantly improves the accuracy and reliability of interferogram analysis.
• The technique shows promise for diverse scientific and engineering applications.