Optics and Precision Engineering, Volume. 31, Issue 22, 3383(2023)

Fine segmentation and stable tracking of spacecraft components

Yadong SHAO1,3,4, Yuanbin SHAO2,3,4, Aodi WU2,3,4, and Xue WAN3,4、*
Author Affiliations
  • 1School of Computer Science and Technology, University of Chinese Academy of Sciences, Beijing00049, China
  • 2School of Aeronautics and Astronautics, University of Chinese Academy of Sciences, Beijing100049, China
  • 3Technology and Engineering Center for Space Utilization, Chinese Academy of Sciences, Beijing100094, China
  • 4Key Laboratory of Space Utilization, Chinese Academy of Sciences, Beijing10009, China
  • show less
    Figures & Tables(16)
    Part of the spacecraft component instance segmentation dataset
    Overall flow chart of the system
    Mask refinement module
    Hierarchical weighted quintuple loss
    Part of the component re-identification dataset
    Schematic representation of segmentation test set and tracking test set
    Segmentation refinement of a solar panel
    Segmentation refinement of a main body
    Training loss curves for each network
    Training loss curves using each loss function
    Component tracking results for short timings
    Component tracking results for long timings
    • Table 1. Result of spacecraft component instance segmentation accuracy on segmentation test set

      View table
      View in Article

      Table 1. Result of spacecraft component instance segmentation accuracy on segmentation test set

      网络模型主干网络Segm_ mAP/%速度/FPS
      Segm_mAP_50:5:95Segm_mAP_75Segm_mAP_s
      SparseInstResNet502057.6069.0061.2023.12
      SparseInst+RMResNet5060.6069.3064.202.82
      SOLOv2ResNet5076.6090.8081.3017.38
      SOLOv2+RMResNet5078.1090.8082.505.62
      YOLACTResNet5078.8093.9081.7034.13
      YOLACT+RMResNet5082.9094.0085.602.07
      Mask RCNNSwinTransformer2279.9091.5093.7017.65
      Mask RCNN+RMSwinTransformer83.6092.3087.701.07
      Cascade Mask RCNNResNet5080.2091.8085.0016.89
      Cascade Mask RCNN+RMResNet5082.0093.0086.901.25
      Mask RCNNResNet5081.8093.9084.6020.55
      Mask RCNN+RMResNet5084.9096.2087.801.54
    • Table 2. Component identification success rate on component identification test set

      View table
      View in Article

      Table 2. Component identification success rate on component identification test set

      方法识别成功率/%
      基于Triplet损失训练的部件重识别网络69.95
      基于Quit_Trihard损失训练的部件重识别网络71.05
      基于分层不加权五元组损失训练的部件重识别网络75.44
      基于分层加权五元组损失训练的部件重识别网络76.86
    • Table 3. Component tracking success rate on tracking test set

      View table
      View in Article

      Table 3. Component tracking success rate on tracking test set

      方法跟踪成功率/%
      基于Triplet损失训练的部件重识别网络进行跟踪85.34
      基于Quit_Trihard损失训练的部件重识别网络进行跟踪86.73
      基于分层不加权五元组损失训练的部件重识别网络进行跟踪78.57
      基于分层加权五元组损失训练的部件重识别网络进行跟踪89.38
    • Table 4. Algorithm running speed

      View table
      View in Article

      Table 4. Algorithm running speed

      模块速度/FPS
      Mask RCNN+RM1.54
      Deep OC SORT202.56
      系统整体1.53
    Tools

    Get Citation

    Copy Citation Text

    Yadong SHAO, Yuanbin SHAO, Aodi WU, Xue WAN. Fine segmentation and stable tracking of spacecraft components[J]. Optics and Precision Engineering, 2023, 31(22): 3383

    Download Citation

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

    Category:

    Received: Jun. 14, 2023

    Accepted: --

    Published Online: Dec. 29, 2023

    The Author Email: Xue WAN (wanxue@csu.ac.cn)

    DOI:10.37188/OPE.20233122.3383

    Topics