Chinese Journal of Lasers, Volume. 50, Issue 19, 1914001(2023)

Deep Learning for Reconstruction of Continuous Terahertz In‐Line Digital Holography

Keyang Cheng* and Qi Li
Author Affiliations
  • State Key Laboratory of Science and Technology on Tunable Laser, Harbin Institute of Technology, Harbin 150080, Heilongjiang, China
  • show less
    Figures & Tables(13)
    Algorithm flow chart of APRA
    Network structure and algorithm block diagram. (a) U-net network structure;(b)H-UnetM block diagram;(c)AS-UnetM block diagram
    Part of label images and input images. (a) Label images;
    Label images, holograms at 19 mm and reconstructed images for character A with 0.4 mm resolution, S with 0.4 mm resolution, A with 0.3 mm resolution, S with 0.3 mm resolution, and G with 0.5 mm resolution from left to right. (a) Label images; (b) holograms;(c)reconstructed images of ASM;(d)reconstructed images of APRA;(e)reconstructed images of H-UnetM;(f)reconstructed images of AS-UnetM
    Holograms at 20 mm and reconstructed images for targets same with Fig. 4. (a) Holograms at 20 mm;(b)reconstructed images of ASM;(c)reconstructed images of APRA;(d)reconstructed images of H-UnetM;(e)reconstructed images of AS-UnetM
    Real illumination intensity distribution
    Holograms and reconstructed images by different methods.(a)Recording distance is 19 mm;(b)recording distance is 20 mm
    Holograms with different frames and corresponding reconstructed images.(a)Single frame hologram;(b)average holograms of 4 frame
    Standard image, holograms and reconstructed images by different methods.(a)Standard image;(b)simulation results;(c)experimental results
    • Table 1. PSNR values of reconstructed images with different methods when z1=19 mm

      View table

      Table 1. PSNR values of reconstructed images with different methods when z1=19 mm

      Methods0.4 mm A0.4 mm S0.3 mm A0.3 mm S0.5 mm G
      ASM65.0364.6167.0567.7763.73
      APRA73.3874.1875.3775.7571.60
      H-UnetM71.5873.0272.6374.2572.83
      AS-UnetM74.0674.9875.0376.8075.95
    • Table 2. PSNRs of reconstructed images with different methods when z1=20 mm

      View table

      Table 2. PSNRs of reconstructed images with different methods when z1=20 mm

      Methods0.4 mm A0.4 mm S0.3 mm A0.3 mm S0.5 mm G
      ASM65.3864.6367.0066.6263.60
      APRA73.3873.7575.4075.8572.10
      H-UnetM71.5672.8472.4273.8372.46
      AS-UnetM74.7275.2775.0476.3976.08
    • Table 3. PSNRs of reconstructed images obtained from average holograms of 40 frame at different recording distances

      View table

      Table 3. PSNRs of reconstructed images obtained from average holograms of 40 frame at different recording distances

      Recording distance /mmPSNR/dB
      ASMAPRAH-UnetMAS-UnetM
      1957.8563.5460.0559.29
      2058.4863.9359.6064.02
    • Table 4. PSNRs of reconstructed images obtained from single frame hologram and average hologram of 4 frame

      View table

      Table 4. PSNRs of reconstructed images obtained from single frame hologram and average hologram of 4 frame

      Hologram typeASMAPRAH-UnetMAS-UnetM
      Single frame56.8063.2158.2654.96
      Average of 4 frame58.1263.8859.8661.40
    Tools

    Get Citation

    Copy Citation Text

    Keyang Cheng, Qi Li. Deep Learning for Reconstruction of Continuous Terahertz In‐Line Digital Holography[J]. Chinese Journal of Lasers, 2023, 50(19): 1914001

    Download Citation

    EndNote(RIS)BibTexPlain Text
    Save article for my favorites
    Paper Information

    Category: terahertz technology

    Received: Aug. 24, 2022

    Accepted: Sep. 28, 2022

    Published Online: Sep. 25, 2023

    The Author Email: Cheng Keyang (c1092986874@163.com)

    DOI:10.3788/CJL221172

    Topics