SE-FSCNet: full-scale connection network for single-shot phase demodulation.

Area of Science:

• Optics and Photonics
• Computer Vision
• Metrology

Background:

• Accurate phase demodulation is crucial for high-precision fringe projection 3D measurement.
• Current deep learning methods, often based on U-Net, face limitations in global information transmission for wrapped phase extraction.
• Improving wrapped phase prediction accuracy is essential for advancing 3D measurement techniques.

Purpose of the Study:

• To propose a novel single-shot phase demodulation method for fringe projection 3D measurement.
• To introduce the SE-FSCNet, a network designed to overcome the global information transmission limitations of U-Net.
• To enhance the accuracy and efficiency of wrapped phase prediction in 3D measurement systems.

Main Methods:

• Development of the SE-FSCNet, a novel full-scale connection network for phase demodulation.
• Implementation of a full-scale connection method and feature fusion module within the decoder for improved feature transmission.
• Integration of a channel attention module (squeeze and excitation) to optimize feature weighting across different scales.

Main Results:

• The SE-FSCNet demonstrates superior feature transmission and utilization capabilities compared to U-Net.
• Ablation studies confirmed the effectiveness of the integrated channel attention module.
• Experimental results show SE-FSCNet achieves higher precision in phase demodulation than traditional Fourier transform and U-Net methods.

Conclusions:

• The proposed SE-FSCNet offers a significant advancement in single-shot phase demodulation for fringe projection 3D measurement.
• The novel network architecture effectively addresses the limitations of existing deep learning models in global information processing.
• SE-FSCNet provides a more accurate and robust solution for 3D measurement applications requiring precise phase unwrapping.