High-fidelity imaging through a perturbed multimode fiber using a u-architecture network with fully connected layers

Area of Science:

• Optics
• Biomedical Engineering
• Artificial Intelligence

Background:

• Multimode fiber (MMF) transmission is susceptible to perturbations, limiting its use in endoscopic imaging.
• Dynamic fiber deformations are common in endoscopic applications, posing significant challenges.

Purpose of the Study:

• To develop a robust imaging method for dynamically perturbed multimode fibers.
• To enhance the quality and efficiency of endoscopic imaging through MMF.

Main Methods:

• A fully connected U-architecture network (FC-UANET) was designed to process speckle patterns.
• The network extracts local and global features to reconstruct high-definition images.
• Comparative experiments evaluated reconstruction accuracy and robustness against perturbations.

Main Results:

• FC-UANET achieved the highest image reconstruction accuracy compared to other methods.
• Accurate image reconstruction from local speckle patterns was demonstrated, improving speed and reducing complexity.
• The network showed generalization ability across different datasets, crucial for endoscopic imaging.

Conclusions:

• FC-UANET offers a robust solution for high-definition endoscopic imaging using dynamically perturbed multimode fibers.
• The proposed method addresses the challenge of fiber deformation in real-world endoscopic scenarios.
• This approach is expected to facilitate the broader application of multimode fibers in endoscopy.