Deep learning-based Phase Measuring Deflectometry for single-shot 3D shape measurement and defect detection of...

Area of Science:

• Computational Imaging
• Optical Metrology
• Deep Learning Applications

Background:

• Phase Measuring Deflectometry (PMD) and Structured-Light Modulation Analysis Technique (SMAT) are effective for specular object measurement.
• Balancing accuracy and speed remains a challenge for these techniques.
• Deep learning has shown promise in computational imaging tasks.

Purpose of the Study:

• To develop a fast and accurate method for specular surface defect detection and 3D shape measurement.
• To leverage deep learning for enhanced performance in PMD and SMAT.
• To recover high-precision modulation and phase distributions from single fringe patterns.

Main Methods:

• An improved U-Net network incorporating depthwise separable convolution and residual structures was developed.
• Deep learning was applied to recover modulation distributions under SMAT from single fringe patterns.
• The method was also adapted to recover phase distributions under PMD for 3D shape measurement.

Main Results:

• The deep learning method successfully recovered high-precision modulation distributions for defect detection under SMAT.
• High-precision phase distributions were recovered for 3D shape measurement under PMD.
• The achieved accuracy closely matched traditional ten-step phase-shifting methods.

Conclusions:

• Deep learning offers a viable solution for fast and accurate specular surface analysis using PMD and SMAT.
• This approach significantly improves upon the speed-accuracy trade-off in existing methods.
• The developed network demonstrates excellent performance in phase and modulation retrieval for specular surfaces.