Optics and Precision Engineering, Volume. 31, Issue 17, 2611(2023)

Lightweight deep learning network for accurate localization of optical image components

Xiaoming NIU1, Li ZENG1、*, Fei YANG2, and Guanghui HE1
Author Affiliations
  • 1College of Mathematics and Statistics, Chongqing University, Chongqing4033, China
  • 2Chang Chun Champion Optics Co., Ltd., Changchun130000, China
  • show less
    Figures & Tables(17)
    Hardware components of localization system
    Optical image localization system for industry component
    Lightweight deep learning localization network architecture
    Flowchart lightweight deep learning localization network algorithm
    Structure of outer network module
    Structure of inner network modules
    Dataset sample
    Trainning loss curves of dataset one
    Trainning loss curves of dataset two
    Localization test results on dataset one
    Localization test results on dataset two
    Localization test results on dataset three
    • Table 1. Model super parameter setting

      View table
      View in Article

      Table 1. Model super parameter setting

      Super parameterValue
      Learning Rate4×10-5
      OptimizerSGD
      Radius5
      Batch size64
      Momentum0.9
    • Table 2. ILWLNet localization test results on dataset one

      View table
      View in Article

      Table 2. ILWLNet localization test results on dataset one

      (Train,Val,Test)Precision/%RMSEMAHDModel
      (10%,10%,80%)95.402.555.11

      57.4k

      (20%,10%,70%)99.201.963.92
      (30%,10%,60%)98.602.725.44
      (40%,10%,50%)99.431.973.93
      (50%,10%,40%)1001.452.90
      (60%,10%,30%)1001.452.91
      (70%,10%,20%)1001.983.96
      (80%,10%,10%)1001.703.40
    • Table 3. ILWLNet localization test results on dataset two

      View table
      View in Article

      Table 3. ILWLNet localization test results on dataset two

      (Train,Val,Test)Precision/%RMSEMAHDModel
      (10%,10%,80%)74.504.128.24

      57.4k

      (20%,10%,70%)98.182.244.49
      (30%,10%,60%)98.901.983.96
      (40%,10%,50%)1001.853.7
      (50%,10%,40%)1001.683.37
      (60%,10%,30%)1002.054.10
      (70%,10%,20%)1001.593.17
      (80%,10%,10%)1001.523.01
    • Table 4. Test results of ILWLNet inference time

      View table
      View in Article

      Table 4. Test results of ILWLNet inference time

      数量

      GPU推理

      平均时间

      CPU阈值分割

      平均时间

      ILWLNet网络

      平均运行时间

      100张8.3486.5894.92
    • Table 5. Comparison of impact of different strategies on model performance

      View table
      View in Article

      Table 5. Comparison of impact of different strategies on model performance

      conv_

      asyn

      Down_

      Block_1

      Down_

      Block_2

      Down_

      Block_3

      Up_

      Block_1

      Up_

      Block_2

      Up_

      Block_3

      FusionWHDPrecisionModel
      87.94%85.1k
      93.62%109k
      92.91%92.2k
      87.23%75.7k
      85.82%54.8k
      100%57.4k
    Tools

    Get Citation

    Copy Citation Text

    Xiaoming NIU, Li ZENG, Fei YANG, Guanghui HE. Lightweight deep learning network for accurate localization of optical image components[J]. Optics and Precision Engineering, 2023, 31(17): 2611

    Download Citation

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

    Category: Information Sciences

    Received: Jun. 5, 2023

    Accepted: --

    Published Online: Oct. 9, 2023

    The Author Email: Li ZENG (drlizeng@cqu.edu.cn)

    DOI:10.37188/OPE.20233117.2611

    Topics