Learnable reconstruction-based synthetic aperture imaging via Fourier ptychography.

Area of Science:

• Optics and Imaging Science
• Computational Imaging
• Artificial Intelligence in Scientific Applications

Background:

• Fourier ptychography (FP) offers super-resolution imaging beyond system limitations.
• Far-field macroscopic FP imaging demands extensive data and suffers from speckle noise, hindering practical use.

Purpose of the Study:

• To develop an efficient method for high-quality Fourier ptychography image reconstruction with reduced data acquisition.
• To address the challenges of limited sampling and speckle noise in macroscopic FP imaging.

Main Methods:

• Introduction of FPADMMNet, a novel model-constrained neural network.
• Integration of the alternating direction method of multipliers (ADMM) optimization algorithm within a neural network framework.
• Mapping ADMM sub-problems to network layers and iterative process to network stages.

Main Results:

• FPADMMNet achieves high-quality image reconstruction even with limited sampling data.
• The proposed method significantly reduces the amount of data required for reconstruction.
• Reconstruction quality is maintained effectively despite data reduction and inherent noise.

Conclusions:

• FPADMMNet offers a powerful solution for overcoming data acquisition and noise limitations in far-field macroscopic Fourier ptychography.
• This AI-driven approach enhances the practicality and efficiency of FP imaging systems.
• The model-constrained network architecture provides a robust framework for advanced computational imaging.