• Infrared and Laser Engineering
  • Vol. 49, Issue 7, 20200170 (2020)
Yunfei Xu1, Duzhou Zhang2、*, Li Wang1, and Baocheng Hua1
Author Affiliations
  • 1北京控制工程研究所 空间光电测量与感知实验室,北京 100190
  • 2中国空间技术研究院,北京 100094
  • show less
    DOI: 10.3788/IRLA20200170 Cite this Article
    Yunfei Xu, Duzhou Zhang, Li Wang, Baocheng Hua. Lightweight feature fusion network design for local feature recognition of non-cooperative target[J]. Infrared and Laser Engineering, 2020, 49(7): 20200170 Copy Citation Text show less
    Design idea of NCDN
    Fig. 1. Design idea of NCDN
    SSD framework based on feature fusion
    Fig. 2. SSD framework based on feature fusion
    Convolutional architecture of MobileNet Depthwise/Pointwise
    Fig. 3. Convolutional architecture of MobileNet Depthwise/Pointwise
    Convolution channel compression
    Fig. 4. Convolution channel compression
    Tango(左)与Mango(右)
    Fig. 5. Tango(左)与Mango(右)
    Local feature definition and label of SPEED
    Fig. 6. Local feature definition and label of SPEED
    Scale changes of target on SPEED
    Fig. 7. Scale changes of target on SPEED
    Comparison effects of NCDN at different distances
    Fig. 8. Comparison effects of NCDN at different distances
    mAP comparison of model changes at different distances
    Fig. 9. mAP comparison of model changes at different distances
    VariablesMeaningVariables’ value/m
    Focus0.0176
    Actual size of Tango
    Pixel size of detector
    Table 1. Meaning of variables in camera model
    Channel compression rateStorage space of model/MBRelative accuracy loss
    19.00%
    10.31%
    5.12%
    2.25%
    Table 2. Influence of channel compression on NCDN
    Yunfei Xu, Duzhou Zhang, Li Wang, Baocheng Hua. Lightweight feature fusion network design for local feature recognition of non-cooperative target[J]. Infrared and Laser Engineering, 2020, 49(7): 20200170
    Download Citation