Optics and Precision Engineering, Volume. 31, Issue 15, 2260(2023)

Light spot detection of diamond wire based on deep learning

Zongqiang FENG, Yipeng YING*, Fujun ZHANG, Yongbo YU, and Yi LIU
Author Affiliations
  • School of Mechanical Engineering, Yanshan University, Qinghuangdao066004, China
  • show less
    Figures & Tables(24)
    Schematic diagram of light spot vision detection system
    Hardware composition of light spot vision detection system
    Prototype of light spot vision inspection system
    Raw images captured by left, center, and right cameras of inspection system
    Partial presentation of blob detection dataset
    Detection results of light spots in Yolox-m model
    Schematic diagram Yolox target detection algorithm
    Schematic diagram of position and width and height of prediction box
    The original enhanced feature extraction network structure
    Lightweight CSP layer structure
    Lightweight and improved Yolox-MobileNetV3 network model structure
    Improved network of attention mechanism
    mAP in target detection of light spot
    mAP value of post-training model
    mAP curve of model training
    F1 curve of model training
    PR curve of model training
    Comparison of detection effect of each model
    • Table 1. Dataset parameter

      View table
      View in Article

      Table 1. Dataset parameter

      原始(张)

      初步挑选后

      (张)

      图像增广后

      (张)

      目标斑点

      (个)

      51 5639 00015 000约105 000
    • Table 2. Experiment hardware configuration and software version parameter

      View table
      View in Article

      Table 2. Experiment hardware configuration and software version parameter

      硬件或软件版本信息参 数
      CPUAMD Ryzen 7 3700x
      内存16G
      GPUNVIDIA GeForce RTX 3060Ti
      操作系统Ubuntu18.04LTS
      CUDA11.3
      CUDNN8.0
      Pytorch1.10.0
      Opencv4.5.4
    • Table 3. Spot detection accuracy and speed of Yolo series

      View table
      View in Article

      Table 3. Spot detection accuracy and speed of Yolo series

      ModelmAP/%Speed/(frame·s-1
      Yolov496.972
      Yolov596.123
      Yolox-m97.283
    • Table 4. Cropped MobileNetv3-Large network structure

      View table
      View in Article

      Table 4. Cropped MobileNetv3-Large network structure

      网络结构输 出
      卷积层,3×3,s=2320×320×16
      倒残差块,3×3,s=1320×320×16
      倒残差块,3×3,s=2160×160×24
      倒残差块,3×3,s=1160×160×24
      倒残差块,5×5,s=180×80×40
      倒残差块,5×5,s=180×80×40
      倒残差块,5×5,s=180×80×40
      倒残差块,3×3,s=240×80×80
      倒残差块,3×3,s=140×80×80
      倒残差块,3×3,s=140×80×80
      倒残差块,3×3,s=140×80×80
      倒残差块,3×3,s=140×40×112
      倒残差块,3×3,s=140×40×112
      倒残差块,5×5,s=220×20×160
      倒残差块,5×5,s=120×20×160
      倒残差块,5×5,s=120×20×160
    • Table 5. Parameters of model experiment

      View table
      View in Article

      Table 5. Parameters of model experiment

      ModelParameters/MSize/MbFLOPs/GmAP/%F1Speed/(frame·s-1
      Yolov4-Tiny5.8722.426.8292.520.92215
      Yolox-Tiny5.0319.207.5794.240.92213
      Yolox-MobileNetV35.3020.237.5194.520.94313
      MCA-Yolox5.6521.587.6495.430.95212
      Yolov464.36245.5371.4097.280.9721
      Yolox25.2896.4436.7596.970.9632
    • Table 5. Model parameter for different backbone feature extraction network replacement

      View table
      View in Article

      Table 5. Model parameter for different backbone feature extraction network replacement

      ModelParameters/MSize/MbFLOPs/GmAP/%
      Yolox-m25.2896.4436.7597.28
      Yolox-MobileNetV35.3020.237.5195.02
      Yolox-ShuffleNetV211.8345.159.6594.43
      Yolox-EfficientNet6.8426.106.1394.04
      Yolox-GhostNet5.0319.206.8792.86
    Tools

    Get Citation

    Copy Citation Text

    Zongqiang FENG, Yipeng YING, Fujun ZHANG, Yongbo YU, Yi LIU. Light spot detection of diamond wire based on deep learning[J]. Optics and Precision Engineering, 2023, 31(15): 2260

    Download Citation

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

    Category: Information Sciences

    Received: Aug. 27, 2022

    Accepted: --

    Published Online: Sep. 5, 2023

    The Author Email: YING Yipeng (yyp2022111@163.com)

    DOI:10.37188/OPE.20233115.2260

    Topics