Deep residual learning for low-order wavefront sensing in high-contrast imaging systems.

Area of Science:

• Astronomy and Astrophysics
• Machine Learning
• Optical Engineering

Background:

• High-contrast astronomical imaging requires precise sensing and correction of low-order wavefront aberrations.
• Current coronagraph systems often use image-based sensing (e.g., Lyot-based low order wavefront sensors - LLOWFS) relying on linear fitting, which has limited dynamic range.

Purpose of the Study:

• To propose and evaluate a deep learning approach using residual learning for non-linear wavefront sensing.
• To overcome the dynamic range limitations of conventional linear wavefront sensing methods in coronagraphy.

Main Methods:

• Utilizing deep neural networks with residual learning techniques for wavefront sensing.
• Applying the deep learning approach to Lyot-based low order wavefront sensors (LLOWFS).

Main Results:

• The deep residual learning approach extends the usable range of LLOWFS by over an order of magnitude compared to conventional methods.
• The method demonstrates robust performance even in low-photon regimes, crucial for exoplanet imaging.

Conclusions:

• Deep learning offers a significant advancement for wavefront sensing in astronomical coronagraphy.
• This non-linear sensing approach improves closed-loop control for systems with large initial wavefront errors and enhances exoplanet detection capabilities.