Optics and Precision Engineering, Volume. 31, Issue 24, 3618(2023)

Fine-grained remote sensing ship open set recognition

Changyuan LIU*... Ting LI and Chaofeng LAN |Show fewer author(s)
Author Affiliations
  • College of Measurement and Control Technology and Communication Engineering, Harbin University of Science and Technology, Harbin150080, China
  • show less
    Figures & Tables(17)
    Open set identification scenario
    Model overall structure block diagram
    Basic architecture of STN
    Structure diagram of parallel convolution module
    Sample example
    Graph of recognition accuracy under different α
    Different δ recognition accuracy curves
    Comparison of STN experiments under different openness
    ResNet34 network and SResNet34 network original image attention diagram
    Experimental comparison of hallucinated inated and reachability under different degrees of openness
    Confusion matrix at 9.25% openness
    • Table 1. Dataset of balanced distribution

      View table
      View in Article

      Table 1. Dataset of balanced distribution

      类别序号类别训练数量
      0油船(Tank ship)400
      1巨型船(Megayacht)400
      2砂船(Sand ship)400
      3货船(Cargo)400
      4集装船(Container)400
      5拖船(Towing ship)400
      6民船(Civil ship)400
      0~6总数2 800
    • Table 2. Datasets of the unbalanced distribution

      View table
      View in Article

      Table 2. Datasets of the unbalanced distribution

      类别序号类别训练数量
      0油船(Tank Ship)220
      1巨型船(Megayacht)180
      2砂船(Sand Ship)250
      3货船(Cargo)300
      4集装船(Container)400
      5拖船(Towing Ship)750
      6民船(Civil Ship)700
      0~6总数2 800
    • Table 3. Test dataset

      View table
      View in Article

      Table 3. Test dataset

      开放度/%未知类包含种类总测试数量
      9.253600
      14.165720
      18.357840
      21.989960
    • Table 4. Comparison on accracy of different data sets

      View table
      View in Article

      Table 4. Comparison on accracy of different data sets

      方法/数据集FGSCR-42FGSC-23
      Resnet44.959.8
      BCNN45.761.3
      OpenMax73.878.8
      OLTR76.280.4
      CenterLoss78.381.6
      本文方法85.187.7
    • Table 5. Comparison of methods on the balanced datasets

      View table
      View in Article

      Table 5. Comparison of methods on the balanced datasets

      方法/开放度9.25%14.16%18.35%21.98%
      AccracyF1-scoreAccracyF1-scoreAccracyF1-scoreAccracyF1-score
      Resnet72.4%75.8%60.3%66.4%50.2%62.5%44.9%60.1%
      BCNN73.6%76.5%61.2%68.2%51.0%64.1%45.7%63.1%
      OpenMax83.7%87.2%78.9%84.0%76.3%81.0%73.8%77.9%
      OLTR88.7%90.8%82.4%86.5%78.5%82.5%76.2%80.1%
      CenterLoss89.0%91.4%83.2%88.2%80.1%84.6%78.3%82.8%
      本文方法90.5%92.2%86.3%89.4%85.7%87.9%85.1%86.4%
    • Table 6. Comparison of methods on the unbalanced datasets

      View table
      View in Article

      Table 6. Comparison of methods on the unbalanced datasets

      方法/开放度9.25%14.16%18.35%21.98%
      AccracyF1-scoreAccracyF1-scoreAccracyF1-scoreAccracyF1-score
      Resnet72.0%75.5%60.3%66.2%50.1%62.4%44.3%59.4%
      BCNN73.3%85.2%61.1%68.0%51.0%63.8%45.5%62.9%
      OpenMax81.6%86.3%75.5%81.2%74.3%78.8%72.7%76.5%
      OLTR85.2%88.1%77.8%83.5%75.5%82.0%74.5%79.8%
      CenterLoss85.5%87.6%78.5%83.1%77.8%82.6%76.5%81.1%
      本文方法90.0%91.9%85.1%88.4%84.3%85.8%84.1%85.3%
    Tools

    Get Citation

    Copy Citation Text

    Changyuan LIU, Ting LI, Chaofeng LAN. Fine-grained remote sensing ship open set recognition[J]. Optics and Precision Engineering, 2023, 31(24): 3618

    Download Citation

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

    Category:

    Received: May. 16, 2023

    Accepted: --

    Published Online: Jan. 5, 2024

    The Author Email: LIU Changyuan (liuchangyuan@hrbust.edu.cn)

    DOI:10.37188/OPE.20233124.3618

    Topics