• Laser & Optoelectronics Progress
  • Vol. 60, Issue 20, 2015004 (2023)
Hao Wang1, Tao Zha1, Lingmei Nie1, Jun Zhang2..., Yuxi Tang2 and Youquan Zhao1,*|Show fewer author(s)
Author Affiliations
  • 1School of Precision Instrument and Opto-Electronics Engineering, Tianjin University, Tianjin 300072, China
  • 2Gansu Constar Technology Group Co., Ltd., Baiyin 730900, Gansu , China
  • show less
    DOI: 10.3788/LOP223132 Cite this Article Set citation alerts
    Hao Wang, Tao Zha, Lingmei Nie, Jun Zhang, Yuxi Tang, Youquan Zhao. Improved Faster R-CNN-Based Contact Lens Surface Defect Detection[J]. Laser & Optoelectronics Progress, 2023, 60(20): 2015004 Copy Citation Text show less
    Structure of Faster R-CNN with FPN
    Fig. 1. Structure of Faster R-CNN with FPN
    Detection platform. (a) Structure of detection platform; (b) physical image of detection platform
    Fig. 2. Detection platform. (a) Structure of detection platform; (b) physical image of detection platform
    Several defects of contact lens. (a) Without defect; (b) bubble; (c) turning point; (d) scratch; (e) mold point; (f) mold edge
    Fig. 3. Several defects of contact lens. (a) Without defect; (b) bubble; (c) turning point; (d) scratch; (e) mold point; (f) mold edge
    Comparison of train loss and precision. (a) Loss function curve; (b) precision curve
    Fig. 4. Comparison of train loss and precision. (a) Loss function curve; (b) precision curve
    K-means++ clustering result
    Fig. 5. K-means++ clustering result
    Results of contact lens defect detection obtained by improved Faster R-CNN algorithm
    Fig. 6. Results of contact lens defect detection obtained by improved Faster R-CNN algorithm
    DefectTotal number of images

    Number pf correct

    detected images

    Number pf false

    detected images

    Number pf missed

    detected images

    Accuracy /%
    Bubble606000100
    Turning point424200100
    Scratch8800100
    Mold point111100100
    Mold edge161600100
    Table 1. Result of defect detection statistics
    ModelBackboneAP /%mAP /%Time /ms
    BubbleTurning pointScratchMold pointMold edge
    Faster R-CNNMobileNetv242.0856.6223.5644.6432.5539.8914.33
    Faster R-CNNVGG1679.6461.558.1463.6377.4458.0825.38
    Faster R-CNNResNet5088.2479.2122.80100.0097.1877.5016.71
    Faster R-CNN+FPNResNet5096.4769.0854.1799.1798.9183.5623.70
    Faster R-CNN+FPN+K-means++ResNet5093.4375.9466.15100.0099.2386.9524.05
    YOLOv3Darknet5382.9151.4430.9382.1171.4663.7711.61
    SSDResNet5083.8747.559.5181.1987.4364.9110.14
    Table 2. Performance comparison of improved Faster R-CNN
    Hao Wang, Tao Zha, Lingmei Nie, Jun Zhang, Yuxi Tang, Youquan Zhao. Improved Faster R-CNN-Based Contact Lens Surface Defect Detection[J]. Laser & Optoelectronics Progress, 2023, 60(20): 2015004
    Download Citation