Optics and Precision Engineering, Volume. 32, Issue 14, 2311(2024)

Iterative reconstruction of compressive sensing combining image hierarchical-feature

Yuhong LIU and Heng YANG*
Author Affiliations
  • School of Electronic and Information Engineering, Lanzhou Jiaotong University, Lanzhou730070, China
  • show less
    Figures & Tables(13)
    Structure of Anchored self-attention
    Structure of overall network
    Multi-scale feature enrichment module
    Secondary spatial-information capture module
    Structure of spatial-hierarchy Transformer
    Comparison of visual effects of different algorithms for reconstructing images at low sampling rates
    Reconfiguration performance with different iterative stages and mixed architectural layers at 0.10 sampling rate
    Variation curves of Loss with training epoch at different sampling rates
    Hierarchical-feature map of SSCM in different iterative stages at 0.25 sampling rate
    • Table 1. Comparison of PSNR(dB)/SSIM of reconstructed images on Set11 dataset with different algorithms for each sampling rate

      View table
      View in Article

      Table 1. Comparison of PSNR(dB)/SSIM of reconstructed images on Set11 dataset with different algorithms for each sampling rate

      算法采样率
      0.010.040.100.250.50平均
      CSNet+20.16/0.517 424.59/0.760 428.29/0.873 933.16/0.942 638.92/0.978 929.02/0.814 6
      AMPNet20.20/0.558 125.26/0.772 229.40/0.877 934.63/0.948 140.33/0.980 429.96/0.827 3
      ISTANet++17.92/0.455 422.88/0.672 328.34/0.853 132.77/0.927 438.73/0.972 728.13/0.776 2
      OPINENet+20.02/0.536 225.52/0.787 929.81/0.890 434.81/0.951 440.19/0.980 030.07/0.829 2
      TransCS20.23/0.504 925.41/0.788 329.54/0.887 735.06/0.954 840.49/0.981 530.15/0.823 4
      CSformer21.58/0.607 526.28/0.806 129.69/0.888 234.80/0.952 440.72/0.982 430.61/0.847 3
      CHFNet21.60/0.608 226.40/0.812 430.10/0.894 935.33/0.956 140.62/0.981 830.81/0.850 7
    • Table 2. Comparison of PSNR(dB)/SSIM of reconstructed images on different datasets with different algorithms for each sampling rate

      View table
      View in Article

      Table 2. Comparison of PSNR(dB)/SSIM of reconstructed images on different datasets with different algorithms for each sampling rate

      数据集算法采样率
      0.010.040.100.250.50平均
      BSD68CSNet+21.76/0.509 024.89/0.676 827.39/0.802 531.03/0.908 335.87/0.967 628.19/0.772 8
      AMPNet22.28/0.538 725.26/0.676 027.86/0.792 631.74/0.904 836.82/0.968 028.79/0.776 0
      ISTANet++19.89/0.462 123.39/0.601 126.39/0.741 629.79/0.862 034.92/0.951 326.88/0.723 6
      OPINENet+21.88/0.516 225.16/0.684 127.81/0.804 031.50/0.906 236.32/0.965 828.53/0.775 3
      TransCS21.55/0.493 325.28/0.688 127.86/0.808 531.74/0.912 136.81/0.969 928.65/0.774 4
      CSformer22.81/0.556 625.73/0.695 428.05/0.803 531.80/0.910 637.13/0.971 629.10/0.787 5
      CHFNet22.94/0.566 125.85/0.702 128.36/0.827 531.96/0.913 936.93/0.970 329.21/0.796 0
      Urban100CSNet+18.47/0.393 221.51/0.630 624.46/0.798 028.75/0.914 634.05/0.970 325.45/0.741 3
      AMPNet18.85/0.434 521.93/0.641 225.28/0.804 930.39/0.925 836.45/0.976 926.58/0.756 7
      ISTANet++16.79/0.347 220.04/0.543 424.48/0.773 228.83/0.894 935.99/0.972 725.23/0.706 3
      OPINENet+18.62/0.418 922.07/0.666 525.77/0.829 730.96/0.935 136.93/0.979 726.87/0.766 0
      TransCS18.32/0.379 522.20/0.675 625.87/0.834 731.14/0.938 737.23/0.981 426.95/0.762 0
      CSformer19.43/0.465 322.62/0.682 325.80/0.826 730.49/0.932 936.45/0.979 326.96/0.777 3
      CHFNet19.50/0.466 922.46/0.681 726.70/0.850 231.80/0.941 837.48/0.984 127.59/0.784 9
    • Table 3. Comparison of the complexity of different algorithms

      View table
      View in Article

      Table 3. Comparison of the complexity of different algorithms

      采样率算法

      网络

      参数量

      平均重构时间(GPU)/s
      Set11BSD68Urban100
      0.10TransCS1.444 M0.706 20.669 80.726 7
      CSformer6.647 M2.772 03.582 34.804 1
      CHFNet0.608 M0.027 30.031 30.029 1
      0.50TransCS2.284 M0.380 00.705 80.711 9
      CSformer6.713 M2.545 63.773 34.632 6
      CHFNet1.028 M0.027 80.027 80.030 5
    • Table 4. Ablation experiments with different strategies at 0.50 sampling rate

      View table
      View in Article

      Table 4. Ablation experiments with different strategies at 0.50 sampling rate

      重构策略测试集的平均PSNR/dB
      MFEMSSCMSet11BSD68Urban100
      全模块区域分支全局分支
      ×40.3436.6637.31
      ×40.4436.4937.24
      ××40.2836.3037.15
      ×39.8035.5636.60
      40.6236.9337.48
    Tools

    Get Citation

    Copy Citation Text

    Yuhong LIU, Heng YANG. Iterative reconstruction of compressive sensing combining image hierarchical-feature[J]. Optics and Precision Engineering, 2024, 32(14): 2311

    Download Citation

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

    Category:

    Received: Mar. 1, 2024

    Accepted: --

    Published Online: Sep. 27, 2024

    The Author Email: YANG Heng (11220697@stu.lzjtu.edu.cn)

    DOI:10.37188/OPE.20243214.2311

    Topics