Abstract:Traditional endoscopes use fiber bundles to transverse the images and have the disadvantages of large size and poor flexibility. Unlike the single mode fibers just guide the fundamental mode, multi-mode fibers (MMFs) accommodate a large number of guided modes. Accordingly, one MMF can be utilized to transverse imaging for endoscopy. This MMF based endoscope can effectively overcome the above shortcomings. However, there are effects such as modal dispersion and mode coupling in the transmission process of multimode fiber, which will cause the image transmitted in the fiber to be distorted or even unrecognizable. The randomness of fiber shape in fabrication detoriosate the distortion. By measuring the transmission matrix, it is possible to recover the image through MMF. But the overall shape and structure of MMF will change in the operation of endoscope. The transmission characteristics of multimode fiber are only closely related to the fiber geometry, but traditional MMF imaging techniques are applicable when the shape of the optical fiber during imaging is consistent with the shape of the optical fiber when the transmission characteristics are calibrated. This work will focus on solving these problems. This paper explores the imaging technology of single multimode optical fiber, proposes the use of neural network to fit the multimode optical fiber transmission matrix, and calculates the input light field through the reverse transmission matrix and the output speckle light field. The traditional real number network has limited fitting effect on the multimode fiber transmission matrix. Inspired by the physical process of multimode fiber modulation of the light field, the real number network is extended to the field of complex numbers, and complex number operations are implemented in the network in the form of real numbers. This article studies the imaging regularity and effectiveness of multimode fiber under the condition of slight deformation and severe deformation. The neural network fitting method of multi-mode optical fiber transmission matrix used in this paper can maintain its effectiveness within a certain deformation range, and has great application potential in some endoscopic applications.

Speckle Classification of a Multimode Fiber Based on Inception V3

Spatially Variant Deblur and Image Enhancement in a Single Multimode Fiber Imaged by Deep Learning

Multimode optical fiber transmission with a deep learning network

Learning to transmit images through optical speckle of a multimode fiber with high fidelity

Research on Image Reconstruction of Multimode Fiber Based on Transfer Learning

Image transmission through a flexible multimode fiber by deep learning

Information distribution on regions of speckle patterns for imaging of multimode fiber

Fiber-Optic Shape Sensing Using Neural Networks Operating on Multispecklegrams

Reconstructing Images Through Multimode Fibers From the Up-Conversion Speckle Patterns via Deep Learning

Deep learning based optical curvature sensor through specklegram detection of multimode fiber

Data-driven Fiber Model Based on the Deep Neural Network with Multi-Head Attention Mechanism

Deep learning sheds new light on non-orthogonal optical multiplexing

Towards optimal multimode fiber imaging by leveraging input polarization and deep learning

Super-resolution imaging through a multimode fiber: the physical upsampling of speckle-driven

All-fiber high-speed image detection enabled by deep learning

High performance few-mode fiber-based light field direction sensing system using deep convolutional neural network: fiber speckle demodulation network (FSDNET)

Research on Single Fiber Imaging Technology under Fiber Deformation

Optical Fiber Distributed Vibration Sensing Using Grayscale Image and Multi-Class Deep Learning Framework for Multi-Event Recognition

Light Spot Mapping Algorithm for Optical Fiber Based on Siamese Network

Neural Network-Based Multimode Fiber Imaging and Characterization Under Thermal Perturbations

Light Propagation Prediction through Multimode Optical Fibers with a Deep Neural Network