Wavefront sensor-less adaptive optics using deep reinforcement learning.

Area of Science:

• Optical microscopy
• Computational imaging
• Machine learning applications

Background:

• Image quality in microscopy is often limited by wavefront aberrations.
• Adaptive optics (AO) systems correct these aberrations to achieve diffraction-limited imaging.
• Traditional AO uses wavefront sensors, while sensor-less adaptive optics (SAO) utilizes image information.

Purpose of the Study:

• To introduce and evaluate a Deep Reinforcement Learning (DRL) approach for sensor-less adaptive optics (SAO).
• To compare the performance of the DRL-based SAO with a conventional Zernike Mode Hill Climbing algorithm.

Main Methods:

• Implementation of a DRL algorithm for SAO control.
• Utilizing a custom-built fluorescence confocal scanning laser microscope for experiments.
• Iterative correction of deformable mirror shapes guided by image information.

Main Results:

• The DRL approach demonstrated superior performance in aberration correction compared to the Zernike Mode Hill Climbing algorithm.
• Enhanced image quality and diffraction-limited performance were achieved using the DRL-SAO system.
• Successful application of DRL for real-time aberration correction in fluorescence microscopy.

Conclusions:

• Deep Reinforcement Learning is a viable and effective method for sensor-less adaptive optics in microscopy.
• DRL-based SAO offers significant advantages over traditional methods for aberration correction.
• This work paves the way for advanced DRL applications in optical imaging systems.