Fiber image transmission technology has been developed for minimal invasive endoscopes

Optical fibers are fundamental components of modern science and technology due to their inherent advantages, providing efficient and secure media for applications such as Internet communication and big data transmission. Compared to single mode fiber (SMF), multimode fiber (MMFS) can support much more guided modes (~103~~104), and the appeal of large capacity information within the hair diameter and image transport It offers the advantages of . This feature places MMFS as an important tool in areas such as quantum information and microendoscopy.

However, MMF poses important challenges. Their highly scattering nature leads to severe modal dispersion during transmission, significantly reducing the quality of transmitted information. Existing technologies such as artificial neural networks (ANNS) and spatial optical modulators (SLMs) have been limited success in reconstructing distorted images after MMF transmission. Despite these advances, direct optical transfer of distortion-free images via MMF using micron-scale integrated optical components remains an elusive goal in optical research.

To address the long-standing challenges of multimode fiber (MMF) transmissions, the research team led by Professor Qiming Zhang and Professor Haoyi Yu of the Faculty of Artificial Intelligence Science and Technology (SAIST) at Shanghai University for Science and Technology (USST) , has introduced a groundbreaking solution. This study has been published in the journal Nature Photonics.

The team integrated a small multilayer optical conversion neural network (DN2S) into the distal end of the MMFS, allowing for full optical image transmission. Free-Space Diffraction Neural Networks (DN2S), considered ONN, directly handles optical matrix growth at the speed of light, and is a more efficient ANN approach based on deep learning that there are many ANN connectivity numbers. It’s been proposed. , optical image classification, decoding, phase detection, etc.

In this work, the researchers adopted a 3D galboscan two-photon nanolithography (GS-TPN) manufacturing approach to integrate with 150 μm small DN2 with a 150 μm x 150 μm 0.35 meters long MMF. Operating within the visible wavelength range, fiber-integrated miniature DN2S based on multilayer diffraction elements, optically infers speckle pattern amplitude and phase information, and directly reconstructs and transmits image through optical fibers. It will come true.

The system achieves a minimum image reconstruction function size of approximately 4.90 μm for a 65 μm x 65 μm real-time input image, achieving average optical intensity contrast and approximately 4%, diffraction efficiency of approximately 4.90 μm, with approximately 4.90 μm 4.90μm has been achieved. 35% per layer. Furthermore, the platform exhibits transfer learning characteristics.

In transmitting 31 HELA cell images not included in the training dataset, the system can maintain high-quality optical image reconstruction and integrate miniaturized DN² with MMF as an unprecedented micron-scale optical inference platform It emphasized sexuality. This innovation paves the way for multifunctional, compact photonic systems.

This study successfully achieved direct optical image transmission via MMF by integrating multilayer optical variable neural network (DN2) with MMFS. Taking advantage of the exceptional computational capabilities of diffraction neural networks, the system is promising for future applications in compact photonic systems and enables wider enhancements.

Integration of compact DN2S with fiber facets will create a new micrometer-scale platform for advancing MMF-based technologies in compact photonic systems, including rigid endoscopes, MMF signal transmission, mode sorting, and short-range quantum optical interconnects. Provided. Furthermore, this approach can be extended to a variety of fiber systems, including single mode, gradient index, and disordered fibers.