• Photonics Research
  • Vol. 11, Issue 12, 2149 (2023)
Xiaoyu Jin1, Jie Zhao1、2、5、*, Dayong Wang1、2、6、*, John J. Healy3、4, Lu Rong1、2, Yunxin Wang1、2, and Shufeng Lin1、2
Author Affiliations
  • 1College of Physics and Optoelectronics, Faculty of Science, Beijing University of Technology, Beijing 100124, China
  • 2Beijing Engineering Research Center of Precision Measurement Technology and Instruments, Beijing 100124, China
  • 3Beijing-Dublin International College, Beijing University of Technology, Beijing 100124, China
  • 4School of Electrical and Electronic Engineering, College of Engineering and Architecture, University College Dublin, Belfield, Dublin 4, Ireland
  • 5e-mail: zhaojie@bjut.edu.cn
  • 6e-mail: wdyong@bjut.edu.cn
  • show less
    DOI: 10.1364/PRJ.493902 Cite this Article Set citation alerts
    Xiaoyu Jin, Jie Zhao, Dayong Wang, John J. Healy, Lu Rong, Yunxin Wang, Shufeng Lin. Continuous-wave terahertz in-line holographic diffraction tomography with the scattering fields reconstructed by a physics-enhanced deep neural network[J]. Photonics Research, 2023, 11(12): 2149 Copy Citation Text show less
    Recording schematic of THz in-line digital hologram at different rotation angles.
    Fig. 1. Recording schematic of THz in-line digital hologram at different rotation angles.
    Flowchart of the PhysenNet to reconstruct the in-line digital hologram. (a) Flowchart of the PhysenNet. (b) Schematic of the U-Net.
    Fig. 2. Flowchart of the PhysenNet to reconstruct the in-line digital hologram. (a) Flowchart of the PhysenNet. (b) Schematic of the U-Net.
    Schematic of the setup of continuous-wave THz in-line digital holography. Off-axis parabolic mirrors, PM1 and PM2; rotational stage, RS.
    Fig. 3. Schematic of the setup of continuous-wave THz in-line digital holography. Off-axis parabolic mirrors, PM1 and PM2; rotational stage, RS.
    Comparison of the reconstructed results of the Siemens star and the cicada wing by different algorithms. (a) Photo of the Siemens star, (b) in-line hologram, (c) preprocessed normalized hologram, and (d)–(h) amplitude distributions by the backpropagation method, the ER method, the IDPR-RI method, the CCTV, and the PhysenNet, respectively.
    Fig. 4. Comparison of the reconstructed results of the Siemens star and the cicada wing by different algorithms. (a) Photo of the Siemens star, (b) in-line hologram, (c) preprocessed normalized hologram, and (d)–(h) amplitude distributions by the backpropagation method, the ER method, the IDPR-RI method, the CCTV, and the PhysenNet, respectively.
    Comparison of the reconstructed results of the cicada wing by different algorithms. (a) Optical photo of the cicada wing; (b) normalized hologram; (c1)–(e1) and (c2)–(e2) amplitude and phase distributions by the IDPR-RI method, the CCTV, and the PhysenNet, respectively; and (f1) and (f2) amplitude and phase profiles of the white dashed line in (c1)–(e1) and (c2)–(e2).
    Fig. 5. Comparison of the reconstructed results of the cicada wing by different algorithms. (a) Optical photo of the cicada wing; (b) normalized hologram; (c1)–(e1) and (c2)–(e2) amplitude and phase distributions by the IDPR-RI method, the CCTV, and the PhysenNet, respectively; and (f1) and (f2) amplitude and phase profiles of the white dashed line in (c1)–(e1) and (c2)–(e2).
    Comparison of the reconstructed results of a PS foam sphere by different algorithms at a single projection angle. (a) Optical photo of the sample, (b) normalized hologram, and (c1)–(g1) and (c2)–(g2) amplitude and phase distributions by the backpropagation method, the ER method, the IDPR-RI method, the CCTV, and the PhysenNet, respectively.
    Fig. 6. Comparison of the reconstructed results of a PS foam sphere by different algorithms at a single projection angle. (a) Optical photo of the sample, (b) normalized hologram, and (c1)–(g1) and (c2)–(g2) amplitude and phase distributions by the backpropagation method, the ER method, the IDPR-RI method, the CCTV, and the PhysenNet, respectively.
    Reconstructed refractive index distribution of a single PS foam sphere by the FBPP method. (a)–(c), (d)–(f), and (g)–(i) Refractive index profiles based on the DT-IDPR-RI, DT-CCTV, and DT-PhysenNet at x–z, y–z, and y–x cross sections, respectively. (j) Refractive index profiles of the white dotted line in (a), (d), and (g) (Visualization 1).
    Fig. 7. Reconstructed refractive index distribution of a single PS foam sphere by the FBPP method. (a)–(c), (d)–(f), and (g)–(i) Refractive index profiles based on the DT-IDPR-RI, DT-CCTV, and DT-PhysenNet at xz, yz, and yx cross sections, respectively. (j) Refractive index profiles of the white dotted line in (a), (d), and (g) (Visualization 1).
    Reconstructed refractive index distribution for two foam spheres. (a)–(c) Refractive index distributions at x–y and x–z (y1=2 mm, y2=3 mm) cross sections. (d) Volume rendering of the 3D refractive index distribution.
    Fig. 8. Reconstructed refractive index distribution for two foam spheres. (a)–(c) Refractive index distributions at xy and xz (y1=2  mm, y2=3  mm) cross sections. (d) Volume rendering of the 3D refractive index distribution.
    AlgorithmsPlatformIterationsTime
    ERCPU2006  s
    IDPR-RICPU5042  s
    CCTVCPU50068  s
    PhysenNetCPU10,000430  min
    PhysenNetCPU + GPU10,0008  min
    Table 1. Comparison of the Runtime for Different Phase Retrieval Algorithms
    Xiaoyu Jin, Jie Zhao, Dayong Wang, John J. Healy, Lu Rong, Yunxin Wang, Shufeng Lin. Continuous-wave terahertz in-line holographic diffraction tomography with the scattering fields reconstructed by a physics-enhanced deep neural network[J]. Photonics Research, 2023, 11(12): 2149
    Download Citation