Single-image super-resolution reconstruction for continuous-wave terahertz imaging systems

Huan Wang; Liying Lang; Yajun Pang; Lei Zhang; Wei Zheng; Sixing Xi

doi:10.3788/IRLA20220292

Journals >Infrared and Laser Engineering >Volume 52 >Issue 1 >Page 20220292 > Article

Infrared and Laser Engineering
Vol. 52, Issue 1, 20220292 (2023)

Single-image super-resolution reconstruction for continuous-wave terahertz imaging systems

Huan Wang^1,2,3, Liying Lang^2,3, Yajun Pang^2,3, Lei Zhang⁴..., Wei Zheng⁵ and Sixing Xi⁴|Show fewer author(s)

Author Affiliations

¹School of Artificial Intelligence, Hebei University of Technology, Tianjin 300401, China

²Center for Advanced Laser Technology, Hebei University of Technology, Tianjin 300401, China

³Hebei Key Laboratory of Advanced Laser Technology and Equipment, Tianjin 300401, China

⁴School of Mathematics & Physics Science and Engineering, Hebei University of Engineering, Handan 075000, China

⁵Department of Physics and Electronic Engineering, Yuncheng University, Yuncheng 044000, China

show less

DOI: 10.3788/IRLA20220292 Cite this Article

Huan Wang, Liying Lang, Yajun Pang, Lei Zhang, Wei Zheng, Sixing Xi. Single-image super-resolution reconstruction for continuous-wave terahertz imaging systems[J]. Infrared and Laser Engineering, 2023, 52(1): 20220292 Copy Citation Text

show less

Fig. 1. Optical path schematic of imaging system

Download full size | View in the Article

Fig. 2. Physical diagram of imaging system

Download full size | View in the Article

Fig. 3. Scanned raw terahertz image

Download full size | View in the Article

Fig. 4. Distribution of the three functions（）三种函数的函数值分布（）

Download full size | View in the Article

Fig. 5. Physical image of the sample. (a) Sample size diagram; (b) Sample location diagram

Download full size | View in the Article

Fig. 6. Reconstructions images. (a) lanczos2 interpolation algorithm; (b) NEDI algorithm; (c) NSL0 algorithm; (d) ICNSL0 algorithm; (e) Detail of lanczos2 interpolation algorithm; (f) Detail of NEDI algorithm; (g) Detail of NSL0 algorithm; (h) Detail of ICNSL0 algorithm

Download full size | View in the Article

输入：图像Y，过完备字典D，递减数列 $ \sigma [{\sigma _1},{\sigma _2},...,{\sigma _n}] $，步长d， $ {r_0} = 0 $输出：稀疏表示系数 $ \alpha $

1. 初始化计算公式 $\alpha = \alpha {}^{\rm{T} }{(\alpha \alpha {}^{\rm{T} })^{ - 1} }{{Y} }$2. 最速下降法取函数一阶导数的相反数，设为阻尼牛顿法初始值 3. 阻尼牛顿法更新优化方向： $ \alpha = \alpha + d $4. 梯度投影更新 $ \alpha $值： $\alpha = \alpha - \alpha {}^{\rm{T} }{(\alpha \alpha {}^{\rm{T} })^{ - 1} }(D\alpha - {{Y} })$5. 计算 $r = {{Y} } - D\alpha$, 若 $ r - {r_0} < \varepsilon $或超出最大迭代次数，输出结果 $ \alpha $；否则跳至步骤3

Table 1. Steps of ICNSL0 algorithm

View in the Article

Measures	lanczos2	NEDI	NSL0	ICNSL0
NIQE	8.648	6.352	7.499	6.481
PSNR	27.713	14.439	22.872 4	21.433
SSIM	0.897 1	0.266	0.304	0.864 2
ES	16.702	20.741	19.596	22.828
AG	1.525	1.961	1.796	2.096

Table 2. NIQE, PSNR, SSIM, edge intensity and average gradient value of the reconstruction results of different algorithms

Download Citation

Save the article for my favorites

Paper Information

微信扫一扫：分享

微信扫一扫：分享