• Journal of Infrared and Millimeter Waves
  • Vol. 40, Issue 4, 554 (2021)
Rui-Min CHEN1、2、3, Shi-Jian LIU1、3、*, Zhuang MIAO1、2、3, and Fan-Ming LI1、3
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
    DOI: 10.11972/j.issn.1001-9014.2021.04.015 Cite this Article
    Rui-Min CHEN, Shi-Jian LIU, Zhuang MIAO, Fan-Ming LI. Infrared aircraft few-shot classification method based on meta learning[J]. Journal of Infrared and Millimeter Waves, 2021, 40(4): 554 Copy Citation Text show less
    Architecture of meta train and meta test
    Fig. 1. Architecture of meta train and meta test
    Model architecture and feature maps of MAML
    Fig. 2. Model architecture and feature maps of MAML
    Comparison with different functions (a) improved with pool & add, (b) improved with conv & add, (c) improved with conv & concat
    Fig. 3. Comparison with different functions (a) improved with pool & add, (b) improved with conv & add, (c) improved with conv & concat
    Model architecture of MLFC
    Fig. 4. Model architecture of MLFC
    Examples of datasets (a)mini-ImageNet dataset,(b)Infra-object dataset
    Fig. 5. Examples of datasets (a)mini-ImageNet dataset,(b)Infra-object dataset
    Test accuracy with different structures (a)5-way 1-shot, (b)5-way 5-shot
    Fig. 6. Test accuracy with different structures (a)5-way 1-shot, (b)5-way 5-shot
    Test accuracy of MAML and ours in different meta-train gradient steps (a)5-way 1-shot, (b)5-way 5-shot
    Fig. 7. Test accuracy of MAML and ours in different meta-train gradient steps (a)5-way 1-shot, (b)5-way 5-shot
    MethodMeta-train gradient steps5-way 1-shot5-way 5-shot
    Accuracy /(%)GFLOPsAccuracy /(%)GFLOPs

    MLFC

    (ours)

    148.77±1.7729.8563.89±0.9249.78
    249.63±1.8950.8863.40±0.9188.54
    349.27±1.8371.9163.92±0.90127.31
    449.34±1.7992.9564.04±0.91166.07
    550.13±1.86113.9864.14±0.90204.83
    MAML11146.56+1.8529.8258.15+0.9549.73
    548.70±1.84113.8763.11±0.92204.63
    Table 1. Accuracy and GFLOPs of different meta-train gradient steps on mini-ImageNet dataset
    MethodPretraining8-way 1-shot8-way 5-shot
    MAMLNo74.62±0.9990.25±0.38
    MLFC (ours)No78.58±0.9791.12±0.37
    MAMLYes76.31±1.0290.72±0.36
    MLFC (ours)Yes81.27±0.9192.74±0.35
    Table 2. Accuracy of different models on Infra-object dataset (%)
    Method8-way 1-shot8-way 5-shot
    Improved Relation Network1378.47±0.9489.82±1.02
    MLFC (ours)81.27±0.9192.74±0.35
    Table 3. Accuracy of different models on Infra-object dataset (%)
    Rui-Min CHEN, Shi-Jian LIU, Zhuang MIAO, Fan-Ming LI. Infrared aircraft few-shot classification method based on meta learning[J]. Journal of Infrared and Millimeter Waves, 2021, 40(4): 554
    Download Citation