• Optics and Precision Engineering
  • Vol. 30, Issue 13, 1631 (2022)
Feng GUO1, Qibing ZHU1,*, Min HUANG1, and Xiaoxiang XU2
Author Affiliations
  • 1Key Laboratory of Advanced Process Control for Light Industry, Ministry of Education, Jiangnan University, Wuxi2422, China
  • 2Wuxi CK Electric Control Equipment Co., Ltd, Wuxi14400, China
  • show less
    DOI: 10.37188/OPE.20223013.1631 Cite this Article
    Feng GUO, Qibing ZHU, Min HUANG, Xiaoxiang XU. Defect detection in ceramic substrate based on improved YOLOV4[J]. Optics and Precision Engineering, 2022, 30(13): 1631 Copy Citation Text show less
    Automatic defect detecting platform for ceramic substrate
    Fig. 1. Automatic defect detecting platform for ceramic substrate
    Image processing
    Fig. 2. Image processing
    Defects of ceramic substrate
    Fig. 3. Defects of ceramic substrate
    Statistical chart of sample number of defects and their size distribution
    Fig. 4. Statistical chart of sample number of defects and their size distribution
    Structure of CCNet
    Fig. 5. Structure of CCNet
    Structure of YOLOV4-CS
    Fig. 6. Structure of YOLOV4-CS
    Convergence plot of loss function
    Fig. 7. Convergence plot of loss function
    Local detection results of different models
    Fig. 8. Local detection results of different models
    Feature map sizeAnchor Size
    19×19(108×121);(20×86);(113×94)
    38×38(51×49);(54×42);(36×53)
    76×76(26×24);(24×18);(22×31)
    152×152(7×9);(11×14);(16×12)
    Table 1. Size of prior boxes
    DefectTotalCorrect detectionFalse detectionMissed detectionPrecision/%Recall/%
    Stain1 0131 0045999.599.1
    Foreign matter222200100100
    Gold edge bulge83833096.5100
    Damage18518215492.497.9
    Ceramic gap74730110098.6
    Total1 3771 364231498.399.0
    Table 2. Results of defect detection
    ModelSizePrecision/%Recall/%Times/s
    Faster R-CNN(600×600)83.382.90.217
    Efficientdet-B3(768×768)81.380.50.256
    YOLOV4(608×608)86.990.10.182
    YOLOV5(640×640)87.789.80.176
    YOLOX(640×640)85.186.50.167
    YOLOV4-CS(608×608)98.399.00.202
    Table 3. Defect detection results of different algorithms
    Small scale branchCCNetGHMCPrecision/%Recall/%
    93.693.3
    90.191.2
    91.995.0
    96.794.2
    96.597.5
    95.697.9
    98.399.0
    Table 4. Results of different ablation experiments
    Attentional mechanismsPrecision/%Recall/%Time/s
    SENet96.495.10.201
    CBAM97.597.60.227
    ECA-Net97.197.40.222
    CCNet98.399.00.202
    Table 5. Results of different kinds of attention networks
    Prior boxPrecision/%Recall/%
    YOLOV496.396.6
    Ours98.399.0
    Table 6. Results of different kinds of prior box
    Feng GUO, Qibing ZHU, Min HUANG, Xiaoxiang XU. Defect detection in ceramic substrate based on improved YOLOV4[J]. Optics and Precision Engineering, 2022, 30(13): 1631
    Download Citation