Abstract:High-fidelity image transmission through multimode fiber is critical for the biomedical imaging and telecommunications industries. However, mode coupling and modal dispersion usually disrupt the distribution of the incident light field and form a speckle pattern at the distal end of the multimode fiber, which makes it challenging to reconstruct the target image. Data-driven deep learning models have been proposed to reconstruct target images from distorted output fields, but the data dependence of the models hinder their practical application. Here, we report a physics-assisted learning-based fiber imaging scheme that overcomes the shortcomings of traditional methods by combining the advantages of the physical model and the learning model. The role of the physical model is to simplify the mapping relationship between the speckle pattern and the target image, thereby reducing the computational complexity. The learning model is applied to learn on an optimized smallscale dataset and achieve high-fidelity image reconstruction according to the optimized direction provided by the physical model. The feasibility of the proposed scheme is verified by imaging experiments, and the results demonstrate that the physics-assisted learning model outperforms the solely data-driven learning model in terms of time cost, experimental effort, and fidelity. The proposed scheme in this work provides an enlightening reference for the combination of model-driven and data-driven methods, and it has the potential to advance the practical application of deep learning in imaging.

Experimental Demonstration of Multimode Fiber Imaging Based on Pre-trained Autoencoder

Image Reconstruction Using Pre-Trained Autoencoder on Multimode Fiber Imaging System

Experimental Demonstration of a Multim Ode Fiber Imaging System Based on Generative Adversarial Networks

Experimental Demonstration of Multimode Fiber Active Illuminated Imaging Via Deep Learning

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

Imaging through a multimode optical fiber with principal component analysis and a variational autoencoder

High-speed Multimode Fiber Imaging System Based on Conditional Generative Adversarial Network

Multimode optical fiber transmission with a deep learning network

High Accuracy Transmission and Recognition of Complex Images Through Multimode Fibers Using Deep Learning

Deep learning-based multimode fiber imaging in multispectral and multipolarimetric channels

Image Transmission Through a Multimode Fiber Based on Transfer Learning

Research on Image Reconstruction of Multimode Fiber Based on Transfer Learning

Deep Learning Image Transmission Through a Multimode Fiber Based on a Small Training Dataset.

Image Reconstruction of Multimode Fiber Scattering Media Based on Deep Learning

Simultaneous Illumination and Imaging Based on a Single Multimode Fiber.

Multimode fiber image reconstruction based on parallel neural network with small training set under wide temperature variations

High-definition Image Transmission Through Dynamically Perturbed Multimode Fiber by a Self-Attention Based Neural Network.

Seeing Through Multimode Fibers with Physics-Assisted Deep Learning

Intensity-only Mode Decomposition on Multimode Fibers using a Densely Connected Convolutional Network

Single-ended Recovery of Optical fiber Transmission Matrices using Neural Networks

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