Enhanced single-frame interferometry via hybrid conv-transformer architecture for ultra-precise phase retrieval

Area of Science:

• Optical metrology
• Image processing
• Artificial intelligence

Background:

• Dynamic single-frame interferometry faces challenges in reconstructing high-quality interferograms for accurate phase retrieval.
• Existing virtual phase shifting techniques struggle with fringe quality and data requirements.

Purpose of the Study:

• To introduce a novel deep learning architecture for high-fidelity interferogram reconstruction in dynamic single-frame interferometry.
• To improve the accuracy and adaptability of phase retrieval across various optical setups.

Main Methods:

• Development of the Transformer Encoder-Convolution Decoder Phase Shift Network (TECD-PSNet).
• Integration of transformer blocks for global context and convolution blocks for feature extraction.
• Implementation of a residual local negative feedback enhancement mechanism for improved detail sensitivity.

Main Results:

• TECD-PSNet achieves high-fidelity interferogram reconstruction.
• Demonstrated 22.9% improvement in Peak Signal-to-Noise Ratio (PSNR) for reconstructed interferograms.
• Achieved 36.7% reduction in Root Mean Square (RMS) error for retrieved phases compared to state-of-the-art methods.
• Reduced the need for extensive training data.

Conclusions:

• TECD-PSNet offers a robust solution for precise dynamic single-frame interferometry.
• The model shows enhanced adaptability to diverse pupil shapes and optical configurations.
• This deep learning approach significantly advances the field of optical metrology.