• Laser & Optoelectronics Progress
  • Vol. 57, Issue 12, 121009 (2020)
Guangshi Zhang1, Guangying Ge1、*, Ronghua Zhu1, and Qun Sun2
Author Affiliations
  • 1College of Physics and Information Engineering, Liaocheng University, Liaocheng, Shandong 252059, China
  • 2College of Mechanical and Automotive Engineering, Liaocheng University, Liaocheng, Shandong 252059, China
  • show less
    DOI: 10.3788/LOP57.121009 Cite this Article Set citation alerts
    Guangshi Zhang, Guangying Ge, Ronghua Zhu, Qun Sun. Gear Defect Detection Based on the Improved YOLOv3 Network[J]. Laser & Optoelectronics Progress, 2020, 57(12): 121009 Copy Citation Text show less
    YOLOv3 network structure
    Fig. 1. YOLOv3 network structure
    Improved network structure
    Fig. 2. Improved network structure
    Loss function and IOU curves of train dataset of YOLOv3_Dense4 network
    Fig. 3. Loss function and IOU curves of train dataset of YOLOv3_Dense4 network
    Loss function and IOU curves of test dataset of YOLOv3_Dense4 network
    Fig. 4. Loss function and IOU curves of test dataset of YOLOv3_Dense4 network
    Defection results of defect gear. (a) Detection results of gears with different defects; (b) detection results of defect gears and flawless gears
    Fig. 5. Defection results of defect gear. (a) Detection results of gears with different defects; (b) detection results of defect gears and flawless gears
    Detection results of gear defects under different light intensities. (a) 249.28 lx; (b) 321.61 lx; (c) 394.93 lx
    Fig. 6. Detection results of gear defects under different light intensities. (a) 249.28 lx; (b) 321.61 lx; (c) 394.93 lx
    NetClassNumber of defectsPTPPFPPFNP /%RmAP /%R /%F1 /%Time /s
    YOLOv3Stain786725216197.1894.5892.2494.650.098
    Miss5724513912591.9878.1584.50
    YOLOv3_Dense3Stain786756103098.6997.6296.1897.420.107
    Miss572475179796.5483.0489.29
    YOLOv3_Dense4Stain78676562199.2298.4597.3398.270.104
    Miss572547132597.6895.6396.64
    Table 1. Performance comparison of different methods
    Light intensity /lxClassNumber of defectsPTPPFPPFNP /%RmAP /%R /%F1 /%
    394.93Stain2562484898.4197.8096.8897.64
    Miss21820861097.2095.4196.30
    321.61Stain2562473998.8098.0096.4897.63
    Miss2182096997.2195.8796.54
    249.28Stain25624131598.7798.2194.1496.40
    Miss21820751197.6494.9596.28
    Table 2. Performance comparison of YOLOv3_Dense4 under different light intensities
    Guangshi Zhang, Guangying Ge, Ronghua Zhu, Qun Sun. Gear Defect Detection Based on the Improved YOLOv3 Network[J]. Laser & Optoelectronics Progress, 2020, 57(12): 121009
    Download Citation