• Acta Photonica Sinica
  • Vol. 52, Issue 8, 0817001 (2023)
Jiajun CHEN1, Kaixiang LI1, Renjian LI1、2, Chunlei SHAO1, Guiye LI1, and Lingling CHEN1、*
Author Affiliations
  • 1College of Health Science and Environmental Engineering,Shenzhen Technology University,Shenzhen 518118,China
  • 2College of Electronic Information Engineering,Beijing University of Aeronautics and Astronautics,Beijing 100191,China
  • show less
    DOI: 10.3788/gzxb20235208.0817001 Cite this Article
    Jiajun CHEN, Kaixiang LI, Renjian LI, Chunlei SHAO, Guiye LI, Lingling CHEN. AtG-DeepLab V3+ Endoscopic Image Enhancement Algorithm Based on Self-attention Mechanism Optimization[J]. Acta Photonica Sinica, 2023, 52(8): 0817001 Copy Citation Text show less
    AtG-DeepLab V3+ model structure based on self-attention mechanism and DeepLab V3+ network
    Fig. 1. AtG-DeepLab V3+ model structure based on self-attention mechanism and DeepLab V3+ network
    Optical endoscopic imaging system diagram based on fiber bundle
    Fig. 2. Optical endoscopic imaging system diagram based on fiber bundle
    Comparison of test target images enhanced by each model
    Fig. 3. Comparison of test target images enhanced by each model
    Comparison of image resolution enhancement by AtG-DeepLab V3+ model
    Fig. 4. Comparison of image resolution enhancement by AtG-DeepLab V3+ model
    Comparison of evaluation indicators of test target images enhanced by each model
    Fig. 5. Comparison of evaluation indicators of test target images enhanced by each model
    Comparison of biological tissue images enhanced by each model
    Fig. 6. Comparison of biological tissue images enhanced by each model
    Comparison of evaluation indicators of biological tissue image enhanced by each model
    Fig. 7. Comparison of evaluation indicators of biological tissue image enhanced by each model
    ParameterValue
    Learning rate0.001
    Beta_10.9
    Beta_20.999
    Momentum0.9
    Batch size2
    Decay0
    Epochs200
    Table 1. Model parameter configuration
    ModelGPU occupancy/MBTraining time/(h/200 epoch)Parameter quantity/MBReconstruction rate/(spiece-1
    U2-Net31 869115160.034 2
    Attention U-Net32 15510.31030.012 8
    GARNN19 75217220.013 7
    DeepLab V3+14 12296960.033 0
    AtG-DeepLab V3+14 5709.27010.031 6
    Table 2. Model training parameter index
    Jiajun CHEN, Kaixiang LI, Renjian LI, Chunlei SHAO, Guiye LI, Lingling CHEN. AtG-DeepLab V3+ Endoscopic Image Enhancement Algorithm Based on Self-attention Mechanism Optimization[J]. Acta Photonica Sinica, 2023, 52(8): 0817001
    Download Citation