Physics-based learning with channel attention for Fourier ptychographic microscopy.

Area of Science:

• Optics
• Computational Imaging
• Machine Learning

Background:

• Fourier ptychographic microscopy (FPM) enables large field-of-view, high-resolution, quantitative phase imaging.
• System errors, including pupil aberration and LED intensity errors, degrade FPM reconstruction quality.
• Current methods struggle to fully correct these system aberrations.

Purpose of the Study:

• To develop an advanced FPM reconstruction method robust to system errors.
• To introduce a channel attention module into physics-based neural networks for adaptive error correction.
• To enhance the performance and practical applicability of FPM.

Main Methods:

• A physics-based neural network incorporating a channel attention module was developed for FPM reconstruction.
• The channel attention module adaptively corrects spatial LED intensity distribution.
• It also synthesizes Zernike modes for pupil function recovery.

Main Results:

• The proposed method demonstrated superior performance in complex field reconstruction compared to state-of-the-art techniques.
• Effective correction of LED intensity errors and accurate recovery of the pupil function were achieved.
• Simulations and experiments validated the method's effectiveness and robustness.

Conclusions:

• The integration of channel attention modules significantly enhances physics-based neural networks for FPM.
• This approach offers improved accuracy in correcting system errors and reconstructing complex fields.
• The method is expected to advance the practical applications of Fourier ptychographic microscopy.