Deep learning-based conjugate orbital angular momentum interferometry for in-plane displacement measurement.

Area of Science:

• Optics and Photonics
• Artificial Intelligence
• Metrology

Background:

• Accurate measurement of in-plane displacements is crucial in various scientific and engineering fields.
• Traditional phase demodulation techniques in interferometry can be complex and sensitive to noise.
• Orbital Angular Momentum (OAM) interferometry offers unique advantages for displacement sensing.

Purpose of the Study:

• To develop a robust and accurate phase demodulation algorithm for in-plane displacement measurement in OAM interferometry.
• To introduce a deep learning-based approach, the Phase Demodulation Hybrid Neural Network (PDHNN), for direct interferogram analysis.
• To enhance the stability and reliability of displacement measurements under noisy conditions.

Main Methods:

• A custom ResNet-transformer architecture was designed for the PDHNN.
• Deformable convolutions and attention mechanisms were incorporated to extract rotation-sensitive features.
• The algorithm was trained and validated using both simulated and experimental petal-shaped interferograms.
• Direct demodulation of interferograms was performed in a single step.

Main Results:

• The PDHNN achieved a demodulation accuracy of 91.60% within a 1° error margin.
• An average displacement error of 0.13 nm was recorded within a 0.1° error range.
• The algorithm demonstrated high robustness and stability in the presence of noise.
• Successful validation was achieved using both simulated and experimental data.

Conclusions:

• The proposed PDHNN algorithm provides an effective solution for accurate in-plane displacement measurement in OAM interferometry.
• Deep learning, particularly the ResNet-transformer architecture, significantly improves phase demodulation robustness and accuracy.
• The method shows great potential for applications requiring precise displacement sensing in challenging environments.