Physics-informed deep learning framework for wavefront sensing via optical beam pattern analysis.

Area of Science:

• Optical Engineering
• Deep Learning Applications
• Wavefront Sensing

Background:

• Non-interferometric wavefront sensing is crucial for optical systems.
• Existing methods often lack accuracy or efficiency.
• Aberrations like spherical aberration, coma, and astigmatism degrade optical performance.

Purpose of the Study:

• To develop a lightweight, physics-informed deep learning framework (PIRNet) for wavefront sensing.
• To enable simultaneous estimation of multiple aberrations from single-shot intensity patterns.
• To enhance accuracy and generalization in optical beam expansion systems.

Main Methods:

• Proposed PIRNet, a deep learning framework integrating physics-informed loss.
• Generated a large dataset using vortex beam propagation simulations and ABCD matrix method.
• Implemented a physics-consistency loss based on optical propagation models.
• Utilized a staged training strategy and learnable uncertainty weighting.

Main Results:

• PIRNet accurately estimates spherical aberration, coma, and astigmatism.
• The physics-consistency loss improves physical plausibility and model generalization.
• PIRNet outperformed ResNet and Xception models in comparative experiments.
• Performance was robust across varying noise levels and cropping ratios.

Conclusions:

• PIRNet offers a promising, accurate, and generalizable approach for wavefront characterization.
• The integration of physical priors enhances deep learning model performance in optics.
• PIRNet has potential applications in adaptive optics and free-space optical communication.