• Laser & Optoelectronics Progress
  • Vol. 61, Issue 8, 0812002 (2024)
Hui Wang1, Jun Wang1,2,*, and Zhaoliang Cao1,2
Author Affiliations
  • 1School of Physical Science and Technology, Suzhou University of Science and Technology, Suzhou 215009, Jiangsu, China
  • 2State Key Laboratory of Applied Optics, Changchun Institute of Optics, Fine Mechanics and Physics, Chinese Academy of Sciences, Changchun 130033, Jilin, China
  • show less
    DOI: 10.3788/LOP231404 Cite this Article Set citation alerts
    Hui Wang, Jun Wang, Zhaoliang Cao. Water Contact Angle Calculation Method Based on Faster RCNN[J]. Laser & Optoelectronics Progress, 2024, 61(8): 0812002 Copy Citation Text show less
    Traditional Faster RCNN model
    Fig. 1. Traditional Faster RCNN model
    ResNet101 architecture model
    Fig. 2. ResNet101 architecture model
    Residual module
    Fig. 3. Residual module
    FPN framework
    Fig. 4. FPN framework
    Network structure of ResNet101 and FPN
    Fig. 5. Network structure of ResNet101 and FPN
    Network framework of channel attention module
    Fig. 6. Network framework of channel attention module
    Spatial attention module network framework
    Fig. 7. Spatial attention module network framework
    CBAM network framework
    Fig. 8. CBAM network framework
    Network structure of improved Faster RCNN
    Fig. 9. Network structure of improved Faster RCNN
    Improved model water droplet recognition effect
    Fig. 10. Improved model water droplet recognition effect
    Water contact angle calculation flow
    Fig. 11. Water contact angle calculation flow
    Canny edge detection
    Fig. 12. Canny edge detection
    Harris corner extraction
    Fig. 13. Harris corner extraction
    IRLS contour fitting
    Fig. 14. IRLS contour fitting
    The right water contact angle calculated by establishing the coordinate system with the center of the ellipse
    Fig. 15. The right water contact angle calculated by establishing the coordinate system with the center of the ellipse
    Comparison of location results of droplet region. (a)‒(f) The detection results of the original Faster RCNN; (g)‒(l) the detection result of improved Faster RCNN
    Fig. 16. Comparison of location results of droplet region. (a)‒(f) The detection results of the original Faster RCNN; (g)‒(l) the detection result of improved Faster RCNN
    Comparison of the original network mAP curve (thin) and the improved Faster RCNN network mAP curve (thick)
    Fig. 17. Comparison of the original network mAP curve (thin) and the improved Faster RCNN network mAP curve (thick)
    Image of contact angle changing over time
    Fig. 18. Image of contact angle changing over time
    Layer nameOutput sizeResNet101Improved ResNet101
    Conv1112×1127×7,64,stide 2
    Conv2_x56×563×3 max pool,stide 2

    1×1,64

    3×3,64×3

    1×1,256

    1×1,64

    2×2,64×3

    1×1,256

    CBAM-block

    Conv3_x28×28

    1×1,128

    3×3,128×4

    1×1,512

    1×1,128

    2×2,128×4

    1×1,512

    CBAM-block

    Conv4_x14×14

    1×1,256

    3×3,256×23

    1×1,1024

    1×1,256

    2×2,256×23

    1×1,1024

    CBAM-block

    Conv5_x7×7

    1×1,512

    3×3,512×3

    1×1,2048

    1×1,512

    2×2,512×3

    1×1,2048

    CBAM-block

    1×1average pool,1000-d FC,Softmax
    Table 1. Network layer parameters before and after ResNet101 improvement
    ModulemAP /%Omission ratio /%
    Basic86.62511.1
    Basic+FPN87.2648.9
    Basic+CBAM87.9127.3
    Basic+Focal loss86.75712.7
    Basic+FPN+CBAM91.5169.4
    Basic+CBAM+Focal loss89.73110.1
    Basic+FPN+Focal loss;89.8627.8
    Basic+CBAM+FPN+Focal loss95.6585.2
    Table 2. Comparison of ablation experiments
    AlgorithmmAP /%Omission ratio /%FPS /(frame/s)
    Faster RCNN(VGG16)84.8648.815
    Faster RCNN(ResNet101)86.0119.619
    SSD80.92414.221
    YOLOv278.32512.130
    Improved Faster RCNN95.6585.226
    Table 3. Comparison results with relevant algorithms
    NumberOursY-LP-FL-S
    θCAθSTDθCAθSTDθCAθSTDθCAθSTD
    176.160.10875.992.03277.151.26575.452.702
    276.020.11077.011.81575.552.65275.461.952
    376.210.10776.451.21375.882.49277.153.120
    476.280.10978.010.96177.255.21675.414.051
    576.030.11277.781.62475.792.92875.241.282
    676.310.10477.082.15276.510.98575.652.882
    776.190.11075.790.91377.403.81577.001.585
    876.350.11277.061.92575.481.98377.314.925
    976.140.11076.421.31374.981.26875.514.462
    1076.330.10877.272.14575.992.31576.452.273
    Table 4. Measurement results of contact angle