Journal of Applied Optics, Volume. 44, Issue 5, 1030(2023)

Classification of combustion state of sintering flame based on CNN-Transformer dual-stream network

Xiuman LIANG... Jinming AN, Xiaohua CAO, Kai ZENG*, Fubin WANG and Hefei LIU |Show fewer author(s)
Author Affiliations
  • College of Electrical Engineering, North China University of Science and Technology, Tangshan 063210, China
  • show less
    Figures & Tables(8)
    CNN-Transformer two-stream network model
    Structure diagram of depth separable convolution
    Cascading feature coupling units
    Three flame state images of sintered section
    • Table 1. Comparison of learning effects of CNN-Transformer network models under different parameters

      View table
      View in Article

      Table 1. Comparison of learning effects of CNN-Transformer network models under different parameters

      OptimizersLearning rateAccuracy rate/%
      Training setValidation set
      SGD0.0199.5695.07
      0.00199.8596.20
      0.000 199.3794.36
      Adam0.0198.6785.69
      0.00199.2392.35
      0.000 198.2678.96
    • Table 2. Comparison of results of different feature fusion methods

      View table
      View in Article

      Table 2. Comparison of results of different feature fusion methods

      MethodsAccuracy/%
      TFN92.53
      LMF93.02
      FCU95.14
      C-FCU96.20
    • Table 3. Ablation experiment of CNN-Transformer algorithm

      View table
      View in Article

      Table 3. Ablation experiment of CNN-Transformer algorithm

      ModelsAccuracy/%Average accuracy/%
      Normal flameOverburning flameUnderburned flame
      CNN93.2592.7291.5692.51
      Transformer95.0093.4593.5594.00
      CNN-Transformer97.0095.3896.2296.20
    • Table 4. Comparison of learning effects between CNN-Transformer dual-stream network model and other models

      View table
      View in Article

      Table 4. Comparison of learning effects between CNN-Transformer dual-stream network model and other models

      ModelsParams/MFLOPs/GAccuracy/%Training time/minSpeed/fs−1
      InceptionV3[14]23.216.0290.5085.7220.79
      ResNet18[5]11.653.8291.0067.9522.53
      ViT[7]55.5077.9190.49157.6512.24
      MobileNet-V2[15]3.510.5887.3534.1530.58
      Conformer[12]23.535.2394.3573.8523.86
      CMT[16]25.104.0295.0076.2920.56
      Ours12.743.5496.2061.2525.78
    Tools

    Get Citation

    Copy Citation Text

    Xiuman LIANG, Jinming AN, Xiaohua CAO, Kai ZENG, Fubin WANG, Hefei LIU. Classification of combustion state of sintering flame based on CNN-Transformer dual-stream network[J]. Journal of Applied Optics, 2023, 44(5): 1030

    Download Citation

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

    Category: Research Articles

    Received: Sep. 14, 2022

    Accepted: --

    Published Online: Mar. 12, 2024

    The Author Email: ZENG Kai (曾凯)

    DOI:10.5768/JAO202344.0502003

    Topics