High-fidelity image transmission and reconstruction through multimode fiber using OAM modes and deep learning

Area of Science:

• Optics and Photonics
• Artificial Intelligence
• Computational Imaging

Background:

• Multimode optical fibers offer high spatial-mode capacity for image transmission.
• Environmental instabilities and complex inputs create speckle, hindering accurate image reconstruction.
• Orbital Angular Momentum (OAM) offers a novel approach to encoding information in light.

Purpose of the Study:

• To develop a high-fidelity image reconstruction method using OAM mode superpositions in multimode optical fibers.
• To enhance image reconstruction accuracy and edge fidelity using deep learning techniques.
• To demonstrate the generalization capabilities of the developed models across different optical conditions.

Main Methods:

• Encoding grayscale images into OAM mode superpositions.
• Utilizing a ResNet-based decoding network with transfer learning for image reconstruction.
• Developing an attention-enhanced DoubleU-Net for reconstructing images with complex edge structures.
• Implementing OAM filtering to improve edge fidelity.

Main Results:

• Achieved up to 99% reconstruction accuracy using the ResNet model with transfer learning, showing generalization across wavelengths.
• The attention-enhanced DoubleU-Net improved reconstruction accuracy by approximately 4% for complex images with rich edge structures, reaching 95% accuracy.
• OAM filtering was experimentally verified to substantially enhance edge fidelity.

Conclusions:

• OAM mode superpositions combined with deep learning provide a robust method for high-fidelity image reconstruction in multimode optical fibers.
• The developed neural network architectures demonstrate strong generalization and improved performance for complex image features.
• This work paves the way for advanced applications in optical information processing, communication, and imaging.