• Laser & Optoelectronics Progress
  • Vol. 59, Issue 10, 1015006 (2022)
Zehui Li1、2, Xindu Chen1、2、*, Jiasheng Huang3, Lei Wu1、2, and Yangqi Lian1、2
Author Affiliations
  • 1Guangdong Provincial Key Laboratory of Computer Integrated Manufacturing, Guangdong University of Technology, Guangzhou 510006, Guangdong , China
  • 2State Key Laboratory of Precision Electronic Manufacturing Technology and Equipment, Guangdong University of Technology, Guangzhou 510006, Guangdong , China
  • 3Cutting Technology Department, Keda Industrial Group Co., Ltd., Foshan 528000, Guangdong , China
  • show less
    DOI: 10.3788/LOP202259.1015006 Cite this Article Set citation alerts
    Zehui Li, Xindu Chen, Jiasheng Huang, Lei Wu, Yangqi Lian. Defect Detection of Texture Tile Using Improved YOLOv3[J]. Laser & Optoelectronics Progress, 2022, 59(10): 1015006 Copy Citation Text show less
    Coordinate transformation of bounding box
    Fig. 1. Coordinate transformation of bounding box
    CAE structure
    Fig. 2. CAE structure
    Training process of the proposed network
    Fig. 3. Training process of the proposed network
    Improved YOLOv3 backbone
    Fig. 4. Improved YOLOv3 backbone
    Analysis of ground truth of the dataset
    Fig. 5. Analysis of ground truth of the dataset
    Images of tile's defect. (a) Pinhole; (b) crack; (c) hole
    Fig. 6. Images of tile's defect. (a) Pinhole; (b) crack; (c) hole
    Results of reconstruction. (a) Source image 1; (b) reconstruction image 1; (c) source image 2; (d) reconstruction image 2
    Fig. 7. Results of reconstruction. (a) Source image 1; (b) reconstruction image 1; (c) source image 2; (d) reconstruction image 2
    Loss-epoch curve
    Fig. 8. Loss-epoch curve
    Comparison of test results. (a) YOLOv3 test result; (b) YOLOv3+CAE test result
    Fig. 9. Comparison of test results. (a) YOLOv3 test result; (b) YOLOv3+CAE test result
    Comparison experiment. (a) Original image with 1500×1500 resolution; (b) brightness disturbance 1; (c) brightness disturbance 2; (d) stitched image
    Fig. 10. Comparison experiment. (a) Original image with 1500×1500 resolution; (b) brightness disturbance 1; (c) brightness disturbance 2; (d) stitched image
    MethodmAP(IOU is 0.5) /%mAP(IOU is 0.95) /%Avg FPS /(frame⋅s-1
    YOLOv378.3759.4811.49
    YOLOv3+CAE83.4968.108.85
    Table 1. Comparison of model performance
    Zehui Li, Xindu Chen, Jiasheng Huang, Lei Wu, Yangqi Lian. Defect Detection of Texture Tile Using Improved YOLOv3[J]. Laser & Optoelectronics Progress, 2022, 59(10): 1015006
    Download Citation