Chinese Journal of Quantum Electronics, Volume. 41, Issue 5, 738(2024)

A method to solve lack of paired data in neutron computed tomography for deep learning by using photon images

GUO Hu1...2, CHEN Shuai2,*, YANG Minghan2, ZHANG Ziheng2,3, SHAO Hui4 and WANG Jianye2 |Show fewer author(s)
Author Affiliations
  • 1Institutes of Physical Science and Information Technology, Anhui University, Hefei 230601, China
  • 2Institute of Nuclear Energy Safety Technology, HFIPS, Chinese Academy of Sciences,Hefei 230031, China
  • 3University of Science and Technology of China, Hefei 230026, China
  • 4Anhui Jianzhu University, Hefei 230022, China
  • show less
    Figures & Tables(10)
    Unsupervised deep reconstruction network composed of encoder, decoder and migration unit
    Images of the experiment sample. (a) Sample perspection of 0°; (b) Sample perspection of 90°;(c) A perspective of the sample
    Reconstruction cross section of neutron CT and photon CT. (a) Neutron CT image;(b) Low-dose photon CT image; (c) Normal-dose photon CT image
    Training loss results for 100 training epochs
    Reconstruction results of the simple structure cross-section of 60% dose neutron CT images by six reconstruction methods. (a) FBP; (b) SART; (c) MDAM; (d) DnCNN; (e) FBPConvNet; (f) REDCNN (a) FBP; (b) SART; (c) MDAM; (d) DnCNN; (e) FBPConvNet; (f) REDCNN
    Local details of the reconstruction of simple structure cross sections of 60% dose neutron CT images by six reconstruction methods. (a) FBP; (b) SART; (c) MDAM; (d) DnCNN; (e) FBPConvNet; (f) REDCNN (a) FBP; (b) SART; (c) MDAM; (d) DnCNN; (e) FBPConvNet; (f) REDCNN
    Local details of the reconstruction of complex structure cross sections of 60% dose neutron CT images by six reconstruction methods. (a) FBP; (b) SART; (c) MDAM; (d) DnCNN; (e) FBPConvNet; (f) REDCNN (a) FBP; (b) SART; (c) MDAM; (d) DnCNN; (e) FBPConvNet; (f) REDCNN
    Reconstruction results of two kinds of structure cross sections of 60% dose neutron CT images by MDAM_skip.(a) Simple structure cross section; (b) Complex structure cross section
    Reconstruction results of two structural cross sections of 40% dose neutron CT images by three reconstruction methods. (a)(d) FBP; (b)(e) SART; (c)(f) MDAM
    • Table 1. Comparison of SSIM, PSNR and RMSE of neutron CT images reconstructed by various methods

      View table
      View in Article

      Table 1. Comparison of SSIM, PSNR and RMSE of neutron CT images reconstructed by various methods

      MethodSSIMPSNRRMSE
      FBPConvNet0.898 ± 0.05436.98 ± 2.860.0127 ± 0.0063
      REDCNN0.863 ± 0.03335.36 ± 2.030.0148 ± 0.0042
      DnCNN0.824 ± 0.04434.76 ± 2.410.0162 ± 0.0063
      SART0.746 ± 0.00430.70 ± 0.097.5014 ± 0.0881
      MDAM0.905 ± 0.01634.90 ± 0.990.0181 ± 0.0025
      MDAM_skip0.864 ± 0.01732.48 ± 0.800.0239 ± 0.0024
    Tools

    Get Citation

    Copy Citation Text

    Hu GUO, Shuai CHEN, Minghan YANG, Ziheng ZHANG, Hui SHAO, Jianye WANG. A method to solve lack of paired data in neutron computed tomography for deep learning by using photon images[J]. Chinese Journal of Quantum Electronics, 2024, 41(5): 738

    Download Citation

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

    Category:

    Received: Nov. 22, 2022

    Accepted: --

    Published Online: Jan. 8, 2025

    The Author Email: CHEN Shuai (shuai.chen@inest.cas.cn)

    DOI:10.3969/j.issn.1007-5461.2024.05.004

    Topics