• Laser & Optoelectronics Progress
  • Vol. 60, Issue 15, 1524001 (2023)
Yi Wang, Xiaojie Gong*, and Jia Cheng
Author Affiliations
  • College of Electrical Engineering, North China University of Science and Technology, Tangshan 063210, Hebei, China
  • show less
    DOI: 10.3788/LOP221756 Cite this Article Set citation alerts
    Yi Wang, Xiaojie Gong, Jia Cheng. Metal Workpiece Surface Defect Segmentation Method Based on Improved U-Net[J]. Laser & Optoelectronics Progress, 2023, 60(15): 1524001 Copy Citation Text show less
    U-net network structure
    Fig. 1. U-net network structure
    Improved U-net network structure
    Fig. 2. Improved U-net network structure
    MAFE structure
    Fig. 3. MAFE structure
    Bottleneck attention module
    Fig. 4. Bottleneck attention module
    Image acquisition visual platform
    Fig. 5. Image acquisition visual platform
    Schematic diagram of cutting and filling. (a) Defect image; (b) image after processing
    Fig. 6. Schematic diagram of cutting and filling. (a) Defect image; (b) image after processing
    Change curves of loss value
    Fig. 7. Change curves of loss value
    Comparison of the segmentation effects. (a) Original images; (b) label; (c) U-net segmentation effect; (d) U-net-MAFE segmentation effect; (e) U-net-BAM segmentation effect; (f) U-net-MAFE-BAM segmentation effect
    Fig. 8. Comparison of the segmentation effects. (a) Original images; (b) label; (c) U-net segmentation effect; (d) U-net-MAFE segmentation effect; (e) U-net-BAM segmentation effect; (f) U-net-MAFE-BAM segmentation effect
    NumberApproachmPAmIOUTest time /s
    1U-net0.84570.82530.156
    2U-net-MAFE0.85370.83650.158
    3U-net-BAM0.85830.84360.161
    4U-net-MAFE-BAM0.87490.86250.165
    Table 1. Data comparison in the experimental network