Journal of Infrared and Millimeter Waves, Volume. 44, Issue 1, 96(2025)

Infrared aircraft few-shot classification method based on cross-correlation network

Zhen HUANG1...2,3, Yong ZHANG1,3,*, and Jin-Fu GONG1,23 |Show fewer author(s)
Author Affiliations
  • 1Key Laboratory of Infrared System Detection and Imaging Technology,Chinese Academy of Sciences,Shanghai 200083,China
  • 2University of Chinese Academy of Sciences,Beijing 100049,China
  • 3Shanghai Institute of Technical Physics,Chinese Academy of Sciences,Shanghai 200083,China
  • show less
    Figures & Tables(10)
    The overall architecture of CCNet model
    Parameter-free self-attention model
    The architecture of cross attention
    (a) Samples of miniImageNet dataset;(b) Samples of miniInfra dataset
    (a) Training and validation accuracy curves of the baseline model and CCNet model on miniImageNet dataset;(b) Training and validation accuracy curves of the baseline model and CCNet model on miniInfra dataset
    Ablation experiment results on miniImageNet and miniInfra dataset
    The class activation mapping (CAM) feature visualization of CCNet
    • Table 1. Classification results on the miniImageNet dataset (average accuracy with 95% confidence interval)

      View table
      View in Article

      Table 1. Classification results on the miniImageNet dataset (average accuracy with 95% confidence interval)

      MethodBackbone5-way 1-shot5-way 5-shot
      CCNet(ours)ResNet1266.20±0.4381.82±0.31
      MAML23ConvNet48.70±0.8463.11±0.92
      RelationNet19ConvNet50.44±0.8265.32±0.70
      CAN24ResNet1263.85±0.4879.44±0.34
      AFHN25ResNet1862.38±0.7278.16±0.56
      PSST26WRN-28-1064.05±0.4980.24±0.45
      NCA27ResNet1262.55±0.1278.27±0.09
      Mata-baseline28ResNet1263.17±0.2379.26±0.17
      MIAN29ResNet1264.27±0.3581.24±0.26
      TFH30ResNet1864.49±0.8479.94±0.60
    • Table 2. Classification results on the miniInfra dataset (average accuracy with 95% confidence interval)

      View table
      View in Article

      Table 2. Classification results on the miniInfra dataset (average accuracy with 95% confidence interval)

      MethodPre-train5-way 1-shot5-way 5-shot8-way 1-shot8-way 5-shot

      Improved

      Relation Network4

      No84.37±1.3193.66±0.7677.56±1.4690.58±0.64
      Yes82.79±0.7594.51±0.8278.47±0.9489.82±1.02
      MLFC2No78.58±0.9791.12±0.37
      Yes81.27±0.9192.74±0.35
      CCNet(ours)No85.58±0.9795.09±0.4681.95±0.6293.26±0.38
    • Table 3. Comparison of accuracy and parameter quantities of different attention modules

      View table
      View in Article

      Table 3. Comparison of accuracy and parameter quantities of different attention modules

      ModuleSelfCrossminiImageNetminiInfraAdd params
      Baseline×64.8681.270 k
      SE31×66.3781.99102.4 k
      SCE32×62.9679.8089.2 k
      LSA33×64.7780.621 644.16 k
      NLSA34×65.6782.34822.1 k
      CBAM35×64.7782.79102.5 k
      SCR9×64.4378.80157.3 k
      CCA9×66.0084.2645.8 k
      SAM×65.8483.310 k
      CA×65.6984.309.41 k
    Tools

    Get Citation

    Copy Citation Text

    Zhen HUANG, Yong ZHANG, Jin-Fu GONG. Infrared aircraft few-shot classification method based on cross-correlation network[J]. Journal of Infrared and Millimeter Waves, 2025, 44(1): 96

    Download Citation

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

    Category: Infrared Optoelectronic System and Application Technology

    Received: Mar. 29, 2024

    Accepted: --

    Published Online: Mar. 5, 2025

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

    DOI:10.11972/j.issn.1001-9014.2025.01.013

    Topics