Unsupervised adaptive coded illumination Fourier ptychographic microscopy based on a physical neural network.

Area of Science:

• Computational imaging
• Microscopy techniques
• Optical physics

Background:

• Fourier Ptychographic Microscopy (FPM) enables large space-bandwidth product imaging by combining low-resolution images with varied illumination.
• Conventional FPM faces limitations in acquisition time and computational load.
• Balancing field of view and resolution remains a key challenge in FPM.

Purpose of the Study:

• To introduce a novel physical neural network for adaptive illumination in FPM.
• To improve the efficiency of FPM acquisition and computation.
• To enhance imaging speed and fidelity in FPM.

Main Methods:

• Development of a physical neural network incorporating temporally-encoded illumination modes as a distinct layer.
• Implementation of adaptive illumination strategies for FPM.
• Direct capture of experimental data using coded illumination patterns with multiple LEDs.

Main Results:

• Validated feasibility and effectiveness through simulations and experiments.
• Achieved direct capture of multiplexed illumination data, bypassing sequential image post-combination.
• Demonstrated state-of-the-art performance in detail fidelity and imaging velocity.

Conclusions:

• The proposed physical neural network significantly enhances FPM efficiency.
• The adaptive illumination method offers a faster and more computationally efficient approach to high-quality imaging.
• This work presents a significant advancement in computational microscopy for improved imaging performance.