Related Experiment Video
Updated: May 5, 2026
06:48
A Multimodal Wide-Field Fourier-Transform Raman Microscope
Published on: December 30, 2025
875
FPM-DCN: Fourier ptychographic microscopy with a dual-constrained network for robust reconstruction in complex
Optics Express
|May 4, 2026
Summary
Fourier ptychographic microscopy (FPM) combined with deep learning faces challenges with large datasets and complex conditions. This study introduces FPM with a dual-constrained network (FPM-DCN) for robust, high-quality imaging with less data.
Area of Science:
- Computational imaging
- Microscopy
- Deep learning applications
Background:
- Fourier ptychographic microscopy (FPM) shows promise but struggles with generalization due to large data needs.
- Deep learning integration in FPM is limited by performance in noisy, aberrated conditions.
Purpose of the Study:
- To develop a robust Fourier ptychographic microscopy method for complex imaging scenarios.
- To enhance FPM reconstruction quality and generalization using a novel deep learning approach.
Main Methods:
- Proposed FPM with a dual-constrained network (FPM-DCN), a weakly-supervised network.
- Integrated an embedded Fourier ptychographic physical model and a non-dense matching perceptual loss.
- Employed a physics-constrained reconstruction network and an aberration correction network with spatial attention.
Main Results:
- Achieved robust, high-contrast, and low-artifact reconstruction under diverse complex imaging conditions.
- Demonstrated enhanced generalization with few-shot training.
- Successfully reduced noise and improved image contrast.
Conclusions:
- FPM-DCN offers a hybrid data-model-driven strategy for superior FPM reconstruction.
- The proposed method overcomes limitations of traditional deep learning approaches in FPM.
- Enables reliable imaging in challenging microscopy environments.
