Opto-Electronic Engineering, Volume. 52, Issue 3, 240296(2025)

Improved weld surface defect detection algorithm from YOLOv8

Runmei Zhang1, Chenfei Pan2, Zihua Chen1、*, Zhong Chen1, and Bin Yuan1
Author Affiliations
  • 1School of Mechanical and Electrical Engineering, Anhui Jianzhu University, Hefei, Anhui 230601, China
  • 2School of Electronic and Information Engineering, Anhui Jianzhu University, Hefei, Anhui 230601, China
  • show less
    Figures & Tables(14)
    YOLOv8 network structure diagram
    GD-YOLO network structure diagram
    DWR module
    GD-CBAM attention module
    Channel attention module
    Angle cost
    Distance cost
    Inner-IoU loss function
    P-R curves before and after improvement
    Comparison of detection effects
    • Table 1. Loss function comparison

      View table
      View in Article

      Table 1. Loss function comparison

      IoU loss functionmAP0.5/%R/%Params/M
      CIoU94.689.32.89
      SIoU92.985.12.89
      GIoU94.088.92.89
      EIoU93.390.32.89
      DIoU93.891.72.89
      WIoU95.089.42.89
      Inner-SIoU95.389.42.89
    • Table 2. Ablation experiments results

      View table
      View in Article

      Table 2. Ablation experiments results

      C2f-DWRCARAFEGSConvInner-SIoUGD-CBAMmAP0.5/%P/%R/%FLOPs/GParams/MFPS
      87.585.581.48.13.01247
      95.191.188.98.02.95286
      94.290.688.78.43.15132
      94.489.990.17.32.79273
      90.989.188.38.13.01259
      95.992.089.511.63.35318
      96.592.389.911.63.35322
      94.691.289.37.42.89243
      95.390.789.47.42.89257
    • Table 3. Results of comparison experiments

      View table
      View in Article

      Table 3. Results of comparison experiments

      ModelmAP0.5/%P/%R/%FLOPs/GParams/MFPS
      YOLOv8n(baseline)87.585.581.48.13.01247
      Faster R-CNN79.883.177.933.341.2095
      YOLOv3-tiny93.493.088.119.012.13242
      YOLOv5n85.784.578.77.12.51282
      YOLOv6n85.084.278.111.84.23285
      YOLOv9t87.485.280.17.61.97229
      YOLOv10n94.388.388.66.52.26292
      RT-DETR90.789.582.9110.232.01117
      YOLOv11n94.289.189.66.32.58307
      Ours95.390.789.47.42.89257
    • Table 4. Experimental results of NEU-DET dataset

      View table
      View in Article

      Table 4. Experimental results of NEU-DET dataset

      ModelmAP0.5/%P/%R/%FLOPs/G
      YOLOv5m74.871.269.048.0
      YOLOv773.766.368.6104.7
      YOLOv8s74.769.169.028.4
      YOLOv8n77.371.970.08.1
      YOLOv9t73.869.770.37.6
      YOLOv11n77.973.974.66.5
      RT-DETR76.772.470.2110.2
      WFRE-YOLOv8s79.473.675.932.6
      Ours78.875.275.77.4
    Tools

    Get Citation

    Copy Citation Text

    Runmei Zhang, Chenfei Pan, Zihua Chen, Zhong Chen, Bin Yuan. Improved weld surface defect detection algorithm from YOLOv8[J]. Opto-Electronic Engineering, 2025, 52(3): 240296

    Download Citation

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

    Category: Article

    Received: Dec. 17, 2024

    Accepted: Feb. 25, 2025

    Published Online: May. 22, 2025

    The Author Email: Zihua Chen (陈梓华)

    DOI:10.12086/oee.2025.240296

    Topics