PUDCN: two-dimensional phase unwrapping with a deformable convolutional network.

Area of Science:

• Optical Imaging and Measurement
• Deep Learning Applications
• Signal Processing

Background:

• Two-dimensional phase unwrapping is crucial for accurate optical imaging and measurement.
• Existing methods face challenges in achieving high accuracy and robustness.

Purpose of the Study:

• To propose a novel deep learning framework, PUDCN, for improved 2D phase unwrapping.
• To enhance feature extraction and phase refinement for more accurate results.

Main Methods:

• Introduction of a novel deep learning framework (PUDCN) for 2D phase unwrapping.
• Integration of deformable convolution and two related plugins for dynamic feature extraction.
• Implementation of a coarse-to-fine strategy for initial unwrapping and subsequent refinement.

Main Results:

• The proposed PUDCN framework demonstrates superior performance compared to existing state-of-the-art methods.
• Successful application of PUDCN to unwrap phases in optical fiber interferometry.
• Validation of the framework's generalization ability across different optical measurement scenarios.

Conclusions:

• PUDCN offers a significant advancement in 2D phase unwrapping accuracy and efficiency.
• The framework's adaptability makes it suitable for complex optical measurement tasks.
• Deep learning, particularly with deformable convolutions, shows great promise for phase unwrapping.