Opto-Electronic Engineering, Volume. 52, Issue 7, 250087(2025)

Optical tissue images segmentation method of ironmaking coke based on MD-UNet

Jinwei Liu1, Huaiguang Liu2,3、*, Ning Ma1, and Pengfei Fu1
Author Affiliations
  • 1School of Mechanical Engineering, Wuhan University of Science and Technology, Wuhan, Hubei 430081, China
  • 2Key Laboratory of Metallurgical Equipment and Control Technology, Ministry of Education, Wuhan, Hubei 430081, China
  • 3Hubei Key Laboratory of Mechanical Transmission and Manufacturing Engineering, Wuhan, Hubei 430081, China
  • show less
    Figures & Tables(21)
    Coke optical tissue microscopy image imaging process
    Characteristics of optical tissue image of coke
    Complexity of optical tissue micrographs of coke
    MD-UNet network framework
    Comparison of dilated convolutional feeling fields with different expansion rates
    Comparison of the structure of the decoder's original convolutional module and the MBDF module
    Comparison plot of the ReLU and GELU activation function
    CloAttention module structure diagram
    Schematic structure of the experimental model
    Experimental results of the CloAttention module. (a) Image group 1; (b) Image group 2; (c) Image group 3
    Feature subgraph change process
    Coke optical tissue segmentation results. (a) Image group 1; (b) Image group 2; (c) Image group 3
    Experimental platforms
    Experimental images and semantic graphs. (a) Image group 1; (b) Image group 2; (c) Image group 3
    Confusion matrix
    • Table 1. Server configurations and their development environment

      View table
      View in Article

      Table 1. Server configurations and their development environment

      ProgramsConfigurations
      Operating systemWindow 10
      CPUIntel (R) Xeon (R) Processor E5-2680 v4
      GPUNVIDIA RTX A4000
      Acceleration moduleCUDA
      CompilerPython 3.7
      Deep learning frameworkPyTorch 1.10
    • Table 2. Experimental results of the CloAttention module

      View table
      View in Article

      Table 2. Experimental results of the CloAttention module

      ModelBackground/%Fiber/%Mottled/%Inert/%mIoU/%F1-Score/%
      Model 192.6681.1689.2889.2588.0993.95
      Model 292.6981.9189.4689.2588.3394.03
      Model 392.5180.8189.0388.8487.8093.72
      Model 492.4981.1089.2289.1687.9993.90
    • Table 3. Ablation experiment model groups

      View table
      View in Article

      Table 3. Ablation experiment model groups

      ModelClarification
      Baseline1U-Net using only VGG16 as backbone network
      Baseline2Baseline1 + CloAttention attention module
      Baseline3Baseline1 + MBDF module
      Baseline4Baseline1 + CloAttention module + MBDF module
    • Table 4. Results of ablation experiments

      View table
      View in Article

      Table 4. Results of ablation experiments

      ModelBackground/%Fiber/%Mottled/%Inert/%mIoU/%F1-Score/%
      Baseline192.6681.1689.2889.2588.0993.95
      Baseline292.6981.9189.4689.2588.3394.03
      Baseline392.6382.5889.8089.4688.6294.20
      Baseline492.5882.9489.9789.4088.7294.28
    • Table 5. Results of comparative experiments

      View table
      View in Article

      Table 5. Results of comparative experiments

      ModelBackground/%Fiber/%Mottled/%Inert/%mIoU/%F1-Score/%
      MD-UNet92.5882.9489.9789.4088.7294.28
      DeepLabv3+91.6779.3987.8087.2186.5293.25
      PSPNet88.5978.1786.2484.0284.2591.77
      SegFormer86.5065.0176.9967.5774.0285.95
    • Table 6. Results of the percentage of each coke's optical tissue composition

      View table
      View in Article

      Table 6. Results of the percentage of each coke's optical tissue composition

      Image codeFiber/%Mottled/%Inert/%
      图14(a)65.8228.825.36
      图14(b)4.1685.7210.12
      图14(c)5.9391.722.35
    Tools

    Get Citation

    Copy Citation Text

    Jinwei Liu, Huaiguang Liu, Ning Ma, Pengfei Fu. Optical tissue images segmentation method of ironmaking coke based on MD-UNet[J]. Opto-Electronic Engineering, 2025, 52(7): 250087

    Download Citation

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

    Category: Article

    Received: Mar. 19, 2025

    Accepted: May. 21, 2025

    Published Online: Sep. 4, 2025

    The Author Email: Huaiguang Liu (刘怀广)

    DOI:10.12086/oee.2025.250087

    Topics