One-shot phase retrieval method for interferometry using a hypercolumns convolutional neural network.

Area of Science:

• Optics and Photonics
• Computer Science
• Metrology

Background:

• Phase retrieval is crucial for 3D profilometry in signal processing.
• Conventional methods require multiple fringe images, limiting efficiency.

Purpose of the Study:

• To propose a novel, single-frame phase retrieval method for interferometry using deep learning.
• To improve the speed and accuracy of phase information extraction in 3D profilometry.

Main Methods:

• A hypercolumns convolutional neural network was designed to treat phase retrieval as a regression problem.
• Training datasets were generated using four mathematical functions, with specific training and validation strategies.
• Polynomial fitting was employed for optimization to correct initial data defects.

Main Results:

• The proposed deep learning method successfully extracts phase information from a single interferogram.
• Experimental validation demonstrated desirable performance in phase retrieval, denoising, and time efficiency.
• A hardware platform based on a point diffraction interferometer was developed to support the method.

Conclusions:

• The developed deep learning-based phase retrieval method offers a significant advancement for 3D profilometry.
• This approach provides a faster and more robust alternative to traditional multi-step phase shift methods.
• The technique shows potential for real-time applications requiring efficient optical measurement.