Optics and Precision Engineering, Volume. 31, Issue 20, 3065(2023)

Defect detection of low-resolution ceramic substrate image based on knowledge distillation

Feng GUO1... Xiaodong SUN1, Qibing ZHU1,*, Min HUANG1 and Xiaoxiang XU2 |Show fewer author(s)
Author Affiliations
  • 1Key Laboratory of Advanced Process Control for Light Industry, Ministry of Education, Jiangnan University, Wuxi 2422, China
  • 2Wuxi CK Electric Control Equipment Co., Ltd, Wuxi 14400, China
  • show less
    Figures & Tables(11)
    Structure of teacher network
    Structure of feature fusion module
    Structure of student network
    Schematic diagram of ceramic substrate and defects
    Schematic diagram of camera field of view
    Local detection results of different models
    Local detection results of different feature fusion methods
    • Table 1. Results of defect detection

      View table
      View in Article

      Table 1. Results of defect detection

      瑕疵类别总数正确检测误检漏检

      Precision

      /%

      Recall

      /%

      合计2 0831 8756220896.8090.01
      污渍1 5911 4413315097.7690.57
      异物28252392.5989.28
      多金34294587.8885.29
      损伤310276153494.8489.03
      缺瓷12010481692.8686.67
    • Table 2. Defect detection results of different algorithms

      View table
      View in Article

      Table 2. Defect detection results of different algorithms

      模型输入尺寸Precision/%Recall/%检测速度/(msf-1
      Faster R-CNN(224×224)--
      Cascade RCNN(224×224)--
      YOLOv4(224×224)85.8675.89126.59
      YOLOv5(224×224)89.1281.2083.02
      YOLOv5(448×448)89.8184.3393.45
      YOLOX(224×224)85.8977.79115.72
      YOLOv6(224×224)88.4479.3587.42
      YOLOv7(224×224)---
      MSAD(224×224)---
      YOLOv4-CS(224×224)93.1285.43142.70
      YOLOv5-CSKD(224×224)96.8090.0187.04
    • Table 3. Experimental results of different feature fusion methods

      View table
      View in Article

      Table 3. Experimental results of different feature fusion methods

      融合方式SizePrecision/%Recall/%
      原始YOLOv5224×22489.1281.20
      SUM224×22495.0188.34
      C-FF224×22495.1388.14
      CAF224×22496.8090.01
    • Table 4. Comparative experimental results using different confidence loss functions

      View table
      View in Article

      Table 4. Comparative experimental results using different confidence loss functions

      教师网络采用GHM-C学生网络采用GHM-C平均准确率/%平均召回率/%
      89.7984.01
      86.5382.33
      86.2288.35
      96.8090.01
    Tools

    Get Citation

    Copy Citation Text

    Feng GUO, Xiaodong SUN, Qibing ZHU, Min HUANG, Xiaoxiang XU. Defect detection of low-resolution ceramic substrate image based on knowledge distillation[J]. Optics and Precision Engineering, 2023, 31(20): 3065

    Download Citation

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

    Category: Information Sciences

    Received: Apr. 10, 2023

    Accepted: --

    Published Online: Nov. 28, 2023

    The Author Email: ZHU Qibing (zhuqib@163.com)

    DOI:10.37188/OPE.20233120.3065

    Topics