• Acta Photonica Sinica
  • Vol. 53, Issue 3, 0312001 (2024)
Chunjia HOU, Boxia HE*, Jinsong HU, Jie YU, and Xuyang CHEN
Author Affiliations
  • School of Mechanical Engineering, Nanjing University of Science and Technology, Nanjing 210094, China
  • show less
    DOI: 10.3788/gzxb20245303.0312001 Cite this Article
    Chunjia HOU, Boxia HE, Jinsong HU, Jie YU, Xuyang CHEN. Surface Defect Detection of Microminiature Aerospace Seals[J]. Acta Photonica Sinica, 2024, 53(3): 0312001 Copy Citation Text show less
    The structure of MMB module
    Fig. 1. The structure of MMB module
    The structure of iRMB module
    Fig. 2. The structure of iRMB module
    The network structure of EMO
    Fig. 3. The network structure of EMO
    The structure of Efficient-FPN Model network and its submodule
    Fig. 4. The structure of Efficient-FPN Model network and its submodule
    The network structure of Next-ViT
    Fig. 5. The network structure of Next-ViT
    The network structure of Transformer-FPN Model
    Fig. 6. The network structure of Transformer-FPN Model
    Type and number of defects in the O-ring datasets
    Fig. 7. Type and number of defects in the O-ring datasets
    The images of O-ring
    Fig. 8. The images of O-ring
    Training results of five detection models
    Fig. 9. Training results of five detection models
    Test results of some images in the test datasets for different models
    Fig. 10. Test results of some images in the test datasets for different models
    ModelNumber of parameters/(×106Amount of computation/(×109Average precision/%Mean average precision/%Detection speed/ (frame·s-1
    Flow markForeign materialParting-line projectionIndentionFlash
    YOLOv5s7.216.668.393.662.871.076.174.4106.1
    YOLOv5x86.8206.373.291.487.973.469.683.364.9
    YOLOv5z51.398.680.494.687.679.181.084.983.3
    Efficient-FPN model2.125.978.494.598.982.474.886.1110.8
    Transformer-FPN model32.797.584.597.499.088.188.091.477.3
    Table 1. Experimental results of five models on O-ring datasets
    ModelDetection time/sAverage detection time/msDetection speed/(frame·s-1
    YOLOv5s15.413.275.9
    YOLOv5x25.221.646.3
    YOLOv5z18.916.261.7
    Efficient-FPN model12.811.091.1
    Transformer-FPN model20.717.856.3
    Table 2. Detection results of the five models on the embedded industrial computer
    Chunjia HOU, Boxia HE, Jinsong HU, Jie YU, Xuyang CHEN. Surface Defect Detection of Microminiature Aerospace Seals[J]. Acta Photonica Sinica, 2024, 53(3): 0312001
    Download Citation