Abstract

In this paper, a new, to the best of our knowledge, neural network combining a new residual neural network (ResNetV2), the residual dense block (RDB), and eHoloNet is proposed to reconstruct a blurred object. With the theory of ghost imaging, only the bucket signal that passes through the blurred object is necessary for reconstruction. The training sets are ENMNIST, which is used for simulation, and the blurred object is designed by Airy convolution. To test the generalization of the neural network, we use multi-slit as the testing sets. Both simulated and experimental results show that the trained neural network is superior in a generalized reconstruction of the blurred object. In addition, the limitation of the reconstruction is also explained in this work.

© 2021 Optical Society of America

Color computational ghost imaging by deep learning based on simulation data training

Zhan Yu, Yang Liu, Jinxi Li, Xing Bai, Zhongzhuo Yang, Yang Ni, and Xin Zhou
Appl. Opt. 61(4) 1022-1029 (2022)

Learning from simulation: An end-to-end deep-learning approach for computational ghost imaging

Fei Wang, Hao Wang, Haichao Wang, Guowei Li, and Guohai Situ
Opt. Express 27(18) 25560-25572 (2019)

Ghost imaging based on asymmetric learning

Tong Bian, Yumeng Dai, Jiale Hu, Zhiyuan Zheng, and Lu Gao
Appl. Opt. 59(30) 9548-9552 (2020)

Relationship between the kernel size of a convolutional layer and the optical point spread function in ghost imaging using deep learning for identifying defect locations

Shoma Kataoka, Yasuhiro Mizutani, Tsutomu Uenohara, Yasuhiro Takaya, and Osamu Matoba
Appl. Opt. 61(23) 6714-6721 (2022)

Single-pixel neural network object classification of sub-Nyquist ghost imaging

Jia-Ning Cao, Yu-Hui Zuo, Hua-Hua Wang, Wei-Dong Feng, Zhi-Xin Yang, Jian Ma, Hao-Ran Du, Lu Gao, and Ze Zhang
Appl. Opt. 60(29) 9180-9187 (2021)