Iterative framework for a high accuracy aberration estimation with one-shot wavefront sensing.

Area of Science:

• Optics and Photonics
• Computational Imaging
• Machine Learning Applications

Background:

• Deep neural networks (DNNs) have advanced image-based wavefront sensing adaptive optics (AO) through non-iterative aberration regression.
• One-shot point spread function (PSF) based algorithms often exhibit limited accuracy in aberration estimation.

Purpose of the Study:

• To develop an iterative closed-loop framework for enhanced wavefront aberration estimation.
• To improve the accuracy of existing non-iterative deep learning methods for AO systems.
• To achieve multi-shot accuracy using a one-shot estimation approach without increased computation or noise sensitivity.

Main Methods:

• An iterative closed-loop framework was proposed for wavefront aberration estimation.
• Defocus point spread functions (PSFs) were simulated for Zernike coefficient estimation.
• A backbone network was trained using the ground-truth defocus PSF and the difference between ground-truth and estimated Zernike coefficients as new labels.
• Model predictions were iteratively updated to refine accuracy.

Main Results:

• The iterative framework demonstrated superior accuracy compared to non-iterative baseline methods with equivalent computational load.
• Experimental results confirmed improved accuracy of existing networks when integrated into the iterative framework.
• The proposed scheme successfully elevated one-shot estimation accuracy to the level of multi-shot methods, even in the presence of noise.

Conclusions:

• The iterative closed-loop framework offers a significant advancement in wavefront aberration estimation for adaptive optics.
• This method enhances the accuracy of DNN-based AO systems, particularly for one-shot PSF approaches.
• The framework provides a computationally efficient pathway to achieve high-accuracy aberration correction comparable to more complex multi-shot techniques.