• Laser & Optoelectronics Progress
  • Vol. 59, Issue 18, 1815008 (2022)
Shisong Zhu1、*, Xiushuai Sun1, Lishan Zhao1, Bibo Lu1, and Donglin Yao2
Author Affiliations
  • 1School of Computer Science and Technology, Henan Polytechnic University, Jiaozuo 454003, Henan , China
  • 2The First Military Representative Office of the Air Force Equipment Department in Wuhan, Wuhan 430000, Hubei , China
  • show less
    DOI: 10.3788/LOP202259.1815008 Cite this Article Set citation alerts
    Shisong Zhu, Xiushuai Sun, Lishan Zhao, Bibo Lu, Donglin Yao. Detection Algorithm of Wire Harness Terminal Core Based on Improved EfficientDet[J]. Laser & Optoelectronics Progress, 2022, 59(18): 1815008 Copy Citation Text show less
    Core detection process of harness terminal microscopic image based on improved EfficientDet model
    Fig. 1. Core detection process of harness terminal microscopic image based on improved EfficientDet model
    Examples of anchor box generation. (a) 8×8 feature map; (b) 4×4 feature map
    Fig. 2. Examples of anchor box generation. (a) 8×8 feature map; (b) 4×4 feature map
    Examples of data augmentation for microscopic images of wire harness terminals. (a) Original photo; (b) flip horizontal; (c) flip vertical; (d) saturation enhancement; (e) hue enhancement; (f) intensity enhancement; (g) salt and pepper noise; (h) Gaussian noise
    Fig. 3. Examples of data augmentation for microscopic images of wire harness terminals. (a) Original photo; (b) flip horizontal; (c) flip vertical; (d) saturation enhancement; (e) hue enhancement; (f) intensity enhancement; (g) salt and pepper noise; (h) Gaussian noise
    Learning rate and loss curves
    Fig. 4. Learning rate and loss curves
    Comparison of detection results of different models
    Fig. 5. Comparison of detection results of different models
    P-R curves and area under the curve. (a) K-means; (b) GHM; (c) K-means and GHM
    Fig. 6. P-R curves and area under the curve. (a) K-means; (b) GHM; (c) K-means and GHM
    Detection results of wire core of micrograph of wire harness terminals. (a) Harness terminal with 37 wire cores; (b) detection result of 37 cores terminal; (c) harness terminal with 84 wire cores; (d) detection result of 84 cores terminal
    Fig. 7. Detection results of wire core of micrograph of wire harness terminals. (a) Harness terminal with 37 wire cores; (b) detection result of 37 cores terminal; (c) harness terminal with 84 wire cores; (d) detection result of 84 cores terminal
    Comparison of counting results between the proposed algorithm and manual method
    Fig. 8. Comparison of counting results between the proposed algorithm and manual method
    DatasetOriginalData augmentationCore of cable
    Train385258277696
    Test96963074
    Total481267880770
    Table 1. Number of images in the harness terminal microscopic image dataset
    Detection frameworkBackboneLoss typemAP /%Elapsed /sSpeed /(frame·s-1
    YOLOv3DarkNet-53CrossEntropy+MSE90.21.7355.61
    RetinaNetResNet-50-FPNFocalLoss+L1Loss89.82.7534.90
    NAS-FPNResNet-50-NAS-FPNFocalLoss+L1Loss89.54.1423.20
    EfficientDet-D0EfficientNet-B0-BiFPNFocalLoss+L1Loss90.71.5761.14
    Improved EfficientDet-D0EfficientNet-B0-BiFPNGHM-C+GHM-R96.21.6458.23
    Table 2. Performance comparison of single-stage target detection model on beam terminal micro image dataset
    Shisong Zhu, Xiushuai Sun, Lishan Zhao, Bibo Lu, Donglin Yao. Detection Algorithm of Wire Harness Terminal Core Based on Improved EfficientDet[J]. Laser & Optoelectronics Progress, 2022, 59(18): 1815008
    Download Citation