Deep Learning Based Image Restoration Method of Optical Synthetic Aperture Imaging System

Ju Tang; Kaiqiang Wang; Wei Zhang; Xiaoyan Wu; Guodong Liu; Jianglei Di; Jianlin Zhao

doi:10.3788/AOS202040.2111001

Acta Optica Sinica, Volume. 40, Issue 21, 2111001(2020)

Deep Learning Based Image Restoration Method of Optical Synthetic Aperture Imaging System

Ju Tang^1,2,3, Kaiqiang Wang^1,2,3, Wei Zhang^1,2,3, Xiaoyan Wu⁴, Guodong Liu⁴, Jianglei Di^1,2,3、*, and Jianlin Zhao^1,2,3、**

Author Affiliations

¹School of Physical Science and Technology, Northwestern Polytechnical University, Xi'an, Shaanxi 710129, China

²Shaanxi Key Laboratory of Optical Information Technology, Xi'an, Shaanxi 710129, China

³Key Laboratory of Material Physics and Chemistry Under Extraordinary Conditions, Ministry of Education, Xi'an, Shaanxi 710129, China

⁴Institute of Fluid Physics, China Academy of Engineering Physics, Mianyang, Sichuan 621900, China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(9)

Fig. 1. Imaging effect of optical synthetic aperture imaging system. (a) Original image; (b) image of synthetic aperture array distribution; (c) imaging map

Download full size

Fig. 2. Comparison of MTF between synthetic aperture system and its equivalent single aperture system. (a) Section curve of main peak of MTF; (b) diagram of MTF of synthetic aperture system; (c) diagram of MTF of equivalent single aperture system

Download full size

Fig. 3. Restoration effect of blind deconvolution method. (a) Restoration maps of blind deconvolution method; (b) ringing phenomenon; (c) comparison of section gray curves

Download full size

Fig. 4. Structure of U-net

Download full size

Fig. 5. Comparison of recovery effects of U-net1 under different training datasets

Download full size

Fig. 6. Comparison of effects of different recovery methods. (a) Original images; (b) imaging maps; (c) images obtained by blind deconvolution; (d) images obtained by U-net2A; (e) images obtained by U-net2B

Download full size

Table 1. Correspondence table of different datasets and networks

View table

Table 1. Correspondence table of different datasets and networks

Experiment Experiment 1			Experiment 2
Dataset	Train set 1	Test set 1	Train set 2A	Train set 2B	Test set 2
Number of maps	2-500	50	1000	1000	300
Number of map types	1	1	2	10	10
Corresponding network	U-net1(N)		U-net2A	U-net2B	U-net2A, U-net2B

Table 2. Average PSNR and SSIM of test set 1
View table
Table 2. Average PSNR and SSIM of test set 1
Parameter Imaging map Restoration map
Blind deconvolution U-net1
(2) (5) (10) (50) (100) (250) (500)
PSNR /dB 16.52 22.99 19.63 20.51 22.21 23.43 23.33 22.73 22.99
SSIM 0.48 0.82 0.64 0.71 0.75 0.78 0.80 0.80 0.81

Table 3. Average PSNR and SSIM of test set 2
View table
Table 3. Average PSNR and SSIM of test set 2
Parameter Imaging map Restoration map
Blind deconvolution U-net2A U-net2B
PSNR /dB 16.08 21.87 20.51 22.21
SSIM 0.39 0.78 0.74 0.77

Tools

Get Citation

Copy Citation Text

Ju Tang, Kaiqiang Wang, Wei Zhang, Xiaoyan Wu, Guodong Liu, Jianglei Di, Jianlin Zhao. Deep Learning Based Image Restoration Method of Optical Synthetic Aperture Imaging System[J]. Acta Optica Sinica, 2020, 40(21): 2111001

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites