Multimode Fiber Image Reconstruction Method Based on Neural Network with Complex-Valued Operation

Langlang Li; Zhen Liu; Mei Zhang; Dong Li; Yang Li; Jiming Ma

doi:10.3788/AOS241873

Acta Optica Sinica, Volume. 45, Issue 7, 0720001(2025)

Multimode Fiber Image Reconstruction Method Based on Neural Network with Complex-Valued Operation

Langlang Li, Zhen Liu^*, Mei Zhang, Dong Li, Yang Li, and Jiming Ma

National Key Laboratory of Intense Pulsed Radiation Simulation and Effect, Northwest Institute of Nuclear Technology, Xi’an 710024, Shaanxi , China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(7)

Fig. 1. Network structure of TMnn model

Download full size

Fig. 2. Examples of training dataset. (a) Original images; (b) speckle images (1-m-long fiber imaging)

Download full size

Fig. 3. Comparison between original images and reconstructed images for a 1-m-long fiber imgaing. (a)(b) Muybridge test dataset; (c)(d) ImageNet validation dataset

Download full size

Fig. 4. Comparison of reconstruction effects of TMnn and traditional neural networks (1-m-long fiber imaging)

Download full size

Fig. 5. Comparison of training processes

Download full size

Table 1. Dataset statistics
View table
Table 1. Dataset statistics
Dataset Size Data sample Label
ImageNet 45000 Digit-speckle images Train
ImageNet 5000 Digit-speckle images Validation
Muybridge 58 Speckle patterns Test

Table 2. Comparison of training information
View table
Table 2. Comparison of training information
Model
Average
SSIM
Time for each epoch /s Actual epochs Total time /h
USINET 0.736 707.06 100 19.64
TMnn 0.740 73.00 650 13.18

Tools

Get Citation

Copy Citation Text

Langlang Li, Zhen Liu, Mei Zhang, Dong Li, Yang Li, Jiming Ma. Multimode Fiber Image Reconstruction Method Based on Neural Network with Complex-Valued Operation[J]. Acta Optica Sinica, 2025, 45(7): 0720001

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites