Image Information Encoding and Decoding Based on Vector Vortex Beam and Deep Learning

Lihu Sun; Pingping Li; Sujuan Liu; Xiaodong Zhang; Nannan Liu; Xinpeng Wu

doi:10.3788/LOP242222

Laser & Optoelectronics Progress, Volume. 62, Issue 10, 1007001(2025)

Image Information Encoding and Decoding Based on Vector Vortex Beam and Deep Learning

Lihu Sun¹, Pingping Li^1,2、*, Sujuan Liu^1,2, Xiaodong Zhang^1,2, Nannan Liu^1,2, and Xinpeng Wu¹

Author Affiliations

¹School of Electronics and Information, Zhengzhou University of Light Industry, Zhengzhou 450001, Henan , China

²Henan Key Laboratory of Magnetoelectronic Information Functional Materials, Zhengzhou University of Light Industry, Zhengzhou 450001, Henan , China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(13)

Fig. 1. Mechanism of grayscale image encoding based on vector vortex beams. (a) Gray-scale values of image pixels were converted to quaternary values; (b) quaternary values were encoded as vector vortex beams; (c) phase holograms for generating left and right rotational circularly polarized vortex beams and intensity distribution of vector vortex beams

Download full size

Fig. 2. Structure of the MRNN model

Download full size

Fig. 3. Internal structure of dense block 1

Download full size

Fig. 4. Perturbation of the optical wavefront phase by atmospheric turbulence of different intensity levels, where the upper right image is the phase screen. (a) Weak turbulence, $C_{n}^{2} = 1 \times 10^{- 16} m^{- 2 / 3}$ ; (b) moderate turbulence, $C_{n}^{2} = 1 \times 10^{- 14} m^{- 2 / 3}$ ; (c) strong turbulence, $C_{n}^{2} = 5 \times 10^{- 13} m^{- 2 / 3}$

Download full size

Fig. 5. Encoding and decoding optical system based on vector vortex beam and MRNN

Download full size

Fig. 6. Polarization distributions of the vector vortex beam, where the rows 1‒4 show the effects of no, weak, moderate, and strong atmospheric turbulence (AT) environments on the spatial polarization distributions of the vector vortex beam, respectively, where the white arrow in the upper-left corner of the figure denotes the transmission direction of the polarizer

Download full size

Fig. 7. Training results of MRNN. (a) Accuracy and loss curves for the validation set in the no-turbulence environment; (b) accuracy curves for the validation set in the weak, moderate and strong turbulence environments; (c)‒(f) confusion matrices for the test set in no, weak, moderate and strong turbulence environments, respectively

Download full size

Fig. 8. Debugging results of each parameter of MRNN model. (a) Effect of input image resolution on the accuracy of MRNN; (b) effect of batch size on the accuracy of MRNN; (c) effect of hyperparameter G on the accuracy of MRNN; (d) effect of pooling layer structure on the accuracy of MRNN

Download full size

Fig. 9. Performance of MRNN versus other neural networks. (a) Connection mechanism of Resnet18; (b) connection mechanism of MRNN; (c) validation accuracy of four deep learning models

Download full size

Fig. 10. Encoding, transmission and decoding experiment of Lena grayscale image

Download full size

Table 1. Parameters of left and right rotationally circularly polarized vortex beams and the synthesized vector vortex beams

View table

Table 1. Parameters of left and right rotationally circularly polarized vortex beams and the synthesized vector vortex beams

$Q_{i} Q_{i}^{'}$	T	m	L_eff	$Q_{i} Q_{i}^{'}$	T	m	L_eff
00	1/16	1	15/17	20	9/16	11	77/25
01	2/16	2	14/9	21	10/16	12	36/13
02	3/16	3	39/19	22	11/16	13	65/27
03	4/16	4	12/5	23	12/16	14	2
10	5/16	6	22/7	30	13/16	16	48/29
11	6/16	7	35/11	31	14/16	17	17/15
12	7/16	8	72/23	32	15/16	18	18/31
13	8/16	9	3	33	16/16	19	0

Table 2. Detailed information of the MRNN model

View table

Table 2. Detailed information of the MRNN model

Layer	Setting	Output shape
Convolution layers	7×7 conv， stride=2	（64， 64， 64）
Pooling	3×3 max pooling， stride=2	（32， 32， 64）
Dense block 1	$[\begin{matrix} 1 \times 1 c o n v \\ 3 \times 3 c o n v \end{matrix}] \times 6$	（32， 32，136）
Transition layer 1	1×1 conv； 2×2 max pooling	（16， 16， 68）
Dense block 2	$[\begin{matrix} 1 \times 1 c o n v \\ 3 \times 3 c o n v \end{matrix}] \times 12$	（16， 16， 212）
Transition layer 2	1×1 conv； 2×2 max pooling	（8， 8， 106）
Dense block 3	$[\begin{matrix} 1 \times 1 c o n v \\ 3 \times 3 c o n v \end{matrix}] \times 24$	（8， 8， 394）
Transition layer 3	1×1 conv； 2×2 max pooling	（4， 4， 197）
Dense block 4	$[\begin{matrix} 1 \times 1 c o n v \\ 3 \times 3 c o n v \end{matrix}] \times 16$	（4， 4， 389）
Classification layer	Global max pooling	（1， 1， 389）
Classification layer	Fully connected+softmax	（16）

Table 3. Epoch corresponding to achieving maximum validation accuracy of the four network models
View table
Table 3. Epoch corresponding to achieving maximum validation accuracy of the four network models
Model Accuracy /% Epoch
Vgg16 99 132
Regnet-x-32gf 98.7 88
Resnet18 99.5 74
MRNN 100 30

Tools

Get Citation

Copy Citation Text

Lihu Sun, Pingping Li, Sujuan Liu, Xiaodong Zhang, Nannan Liu, Xinpeng Wu. Image Information Encoding and Decoding Based on Vector Vortex Beam and Deep Learning[J]. Laser & Optoelectronics Progress, 2025, 62(10): 1007001

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites

Paper Information

Category: Fourier Optics and Signal Processing

Received: Nov. 5, 2024

Accepted: Nov. 26, 2024

Published Online: Apr. 23, 2025

The Author Email: Pingping Li (2019040@zzuli.edu.cn)

DOI:10.3788/LOP242222

CSTR:32186.14.LOP242222

Topics

laser devices and laser physics

Lasers and Laser Optics

Laser physics

laser manufacturing

Instrumentation, Measurement and Metrology

Table 1. Parameters of left and right rotationally circularly polarized vortex beams and the synthesized vector vortex beams

Table 1. Parameters of left and right rotationally circularly polarized vortex beams and the synthesized vector vortex beams

Table 2. Detailed information of the MRNN model

Table 2. Detailed information of the MRNN model

Table 3. Epoch corresponding to achieving maximum validation accuracy of the four network models

Table 3. Epoch corresponding to achieving maximum validation accuracy of the four network models