One-step robust deep learning phase unwrapping.

Area of Science:

• Optics and Photonics
• Artificial Intelligence
• Biomedical Imaging

Background:

• Phase unwrapping is critical for accurate phase measurement but remains challenging, particularly with noise and aliasing.
• Existing methods struggle with complex, real-world phase data, limiting applications in fields like microscopy and fluid dynamics.

Purpose of the Study:

• To develop a robust, one-step deep learning method for phase unwrapping.
• To address limitations of classical phase unwrapping techniques in noisy and aliased conditions.
• To demonstrate the method's effectiveness on diverse, complex phase fields.

Main Methods:

• A database generation technique for phase-type objects was created.
• A one-step deep learning model utilizing a trained deep neural network was implemented.
• The method was tested on unseen phase fields from living mouse osteoblasts and dynamic candle flames.

Main Results:

• The deep neural network successfully performed one-step phase unwrapping on complex, unseen phase fields.
• The method demonstrated excellent performance in resisting noise and aliasing.
• Results showed superior performance compared to classical phase unwrapping methods.

Conclusions:

• Deep learning offers a powerful, direct solution for complex phase unwrapping tasks.
• The proposed method provides a robust and efficient alternative to traditional phase unwrapping algorithms.
• This approach has significant potential for applications requiring precise phase measurement under adverse conditions.