One step accurate phase demodulation from a closed fringe pattern with the convolutional neural network HRUnet

Area of Science:

• Optical Interferometry
• Image Processing
• Machine Learning

Background:

• Phase map retrieval from single closed fringe patterns is a significant challenge in optical interferometry.
• Existing methods often struggle with accuracy and efficiency.

Purpose of the Study:

• To propose a novel convolutional neural network (CNN), HRUnet, for accurate phase demodulation from closed fringe patterns.
• To demonstrate the effectiveness of HRUnet using simulated and real fringe pattern data.

Main Methods:

• Developed HRUnet, a CNN derived from the Unet model, incorporating a high-resolution network (HRnet) module and residual blocks.
• Trained the network to directly output unwrapped phase maps from scaled fringe patterns.
• Compared HRUnet's performance against two other CNNs.

Main Results:

• HRUnet successfully demodulated phase from both simulated and actual fringe patterns with high accuracy.
• The proposed HRUnet demonstrated superior accuracy compared to two other contemporary CNN models.
• The network effectively extracts high-resolution feature maps and mitigates gradient vanishing.

Conclusions:

• HRUnet offers a highly accurate and effective solution for phase map retrieval from single closed fringe patterns.
• The integration of HRnet modules and residual blocks significantly enhances CNN performance in this application.
• This approach advances phase demodulation techniques in optical interferometry.