Journal of Infrared and Millimeter Waves, Volume. 40, Issue 6, 858(2021)

Real-time infrared target detection based on center points

Zhuang MIAO1,2, Yong ZHANG1、*, and Wei-Hua LI1,2
Author Affiliations
  • 1Key Laboratory of Infrared System Detection and Imaging Technology,Shanghai Institute of Technical Physics,Chinese Academy of Sciences,Shanghai 200083,China
  • 2School of Electronic,Electrical,and Communication Engineering,University of Chinese Academy of Sciences,Beijing 100049,China
  • show less
    Figures & Tables(13)
    The overall architecture of TCPD
    The structure of blocks in FEM
    The network structure of FFM
    The network structure of BSM
    The network structure of TPM
    Examples on infrared dataset
    Examples on the VOC dataset
    • Table 1. Network structure of FEM

      View table
      View in Article

      Table 1. Network structure of FEM

      StageOutput SizeOutput ChannelsLayer
      Input384×3843Image
      Stage1192×192243×3,Conv,s2
      Stage296×96243×3,Max Pooling,s2
      Stage348×48116

      Block1×1

      Block2×4

      Stage424×24232

      Block1×1

      Block2×8

      Stage512×12464

      Block1×1

      Block2×4

    • Table 2. Detection results on infrared dataset

      View table
      View in Article

      Table 2. Detection results on infrared dataset

      ModelInput SizemAP/(%)AP /(%)
      BirdFighterAirlinerHelicopterTrainer
      CenterNet384×38488.0476.7388.9594.9190.7788.84
      YOLOv3416×41693.0287.7093.9795.9794.8492.66
      Tiny-YOLOv3416×41680.0866.5883.1693.8584.9271.90
      Tiny-YOLOv4512×51282.8785.6091.0695.3589.1353.23
      FKPD384×38488.9879.4090.8495.0190.2789.39
      TCPD384×38490.2479.4490.6996.0294.6890.35
    • Table 3. Detection results on VOC dataset

      View table
      View in Article

      Table 3. Detection results on VOC dataset

      ModelInput SizemAP/(%)
      CenterNet384×38468.24
      YOLOv3416×41676.80
      Tiny-YOLOv3416×41658.40
      Tiny-YOLOv4416×41665.71
      FKPD384×38461.61
      TCPD384×38466.76
    • Table 4. Real-time analysis of TCPD

      View table
      View in Article

      Table 4. Real-time analysis of TCPD

      ModelFLOPs/BnParameters/MInference Time/ms
      CenterNet8.6914.2248.90
      YOLOv327.9361.63134.07
      Tiny-YOLOv32.348.6831.71
      Tiny-YOLOv42.915.8826.23
      FKPD1.552.0325.86
      TCPD0.490.9521.69
    • Table 5. Ablation study on the design of model

      View table
      View in Article

      Table 5. Ablation study on the design of model

      ModelInput SizemAP/(%)Inference Time/ms
      TCPD(baseline)384×38490.2421.69
      TCPD-small320×32089.8518.22
      TCPD-large512×51292.3832.70
      TCPD-compressed384×38488.6017.90
      TCPD w/o FFM384×38489.2920.15
      TCPD w/o BSM384×38488.7520.43
    • Table 6. Ablation study of Gaussian kernel

      View table
      View in Article

      Table 6. Ablation study of Gaussian kernel

      αDataset (mAP/(%))
      InfraredVOC
      0.3589.4264.81
      0.5590.5666.00
      0.7590.2466.76
      0.9590.3166.24
    Tools

    Get Citation

    Copy Citation Text

    Zhuang MIAO, Yong ZHANG, Wei-Hua LI. Real-time infrared target detection based on center points[J]. Journal of Infrared and Millimeter Waves, 2021, 40(6): 858

    Download Citation

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

    Category: Research Articles

    Received: Feb. 16, 2021

    Accepted: --

    Published Online: Feb. 16, 2022

    The Author Email: Yong ZHANG (zybxy@sina.com)

    DOI:10.11972/j.issn.1001-9014.2021.06.021

    Topics