• Acta Photonica Sinica
  • Vol. 51, Issue 2, 0210001 (2022)
Zijian ZHU1, Qi LIU2, Hongfen CHEN1, Guiyang ZHANG3, Fukuan WANG4, and Ju HUO2、*
Author Affiliations
  • 1School of Astronautics,Harbin Institute of Technology,Harbin 150001,China
  • 2School of Electrical Engineering and Automation,Harbin Institute of Technology,Harbin 150001,China
  • 3School of Mechanical Engineering,Changshu Institute of Technology,Suzhou,Jiangsu 215500,China
  • 4School of Mechanical Engineering,Guangxi University,Nanning 530004,China
  • show less
    DOI: 10.3788/gzxb20225102.0210001 Cite this Article
    Zijian ZHU, Qi LIU, Hongfen CHEN, Guiyang ZHANG, Fukuan WANG, Ju HUO. Infrared Small Vehicle Detection Based on Parallel Fusion Network[J]. Acta Photonica Sinica, 2022, 51(2): 0210001 Copy Citation Text show less
    Residual block
    Fig. 1. Residual block
    The backbone and feature map of YOLOv3
    Fig. 2. The backbone and feature map of YOLOv3
    PaRNet structure diagram
    Fig. 3. PaRNet structure diagram
    Parallel residual block designing diagram
    Fig. 4. Parallel residual block designing diagram
    Schematic diagram of the receptive field of channel fusion strategy
    Fig. 5. Schematic diagram of the receptive field of channel fusion strategy
    Architecture of PaRNet-35 and PaRNet-51
    Fig. 6. Architecture of PaRNet-35 and PaRNet-51
    Deconvolution and upsampling
    Fig. 7. Deconvolution and upsampling
    Diagram of feature fusion based on cross-layer connection
    Fig. 8. Diagram of feature fusion based on cross-layer connection
    The change of TOP-1 value in different networks with the training epochs on CIFAR-10
    Fig. 9. The change of TOP-1 value in different networks with the training epochs on CIFAR-10
    The change of TOP-1 value in different networks with the training epochs on CIFAR-100
    Fig. 10. The change of TOP-1 value in different networks with the training epochs on CIFAR-100
    Part of the frame image of the meeting
    Fig. 11. Part of the frame image of the meeting
    Part of the frame image of the multiple fuzzy targets
    Fig. 12. Part of the frame image of the multiple fuzzy targets
    Detection visualization
    Fig. 13. Detection visualization
    Params/(×106FLOPs/(×108TOP-1/%TOP-5/%
    GoogLeNet6.63.913.010.72
    ResNet-5025.61118.321.29
    ResNet-10144.52016.711.04
    PaRNet-3522.84.414.190.67
    PaRNet-5131.71013.380.64
    Table 1. Performance of different networks on CIFAR-10
    Params/(×106FLOPs/(×108TOP-1/%TOP-5/%
    GoogLeNet6.63.941.9717.27
    ResNet-5025.61148.5022.81
    ResNet-10144.52048.0922.00
    PaRNet-3522.84.441.0817.19
    PaRNet-5131.71039.6715.86
    Table 2. Performance of different networks on CIFAR-100
    Scene sequenceFAR/%MDR/%FPS
    YOLOv3FCR-GOursYOLOv3FCR-GOursYOLOv3FCR-GOurs
    Seq 1:Platform movement0.020.010.010.211.170.1533.91.133.1
    Seq 2:Target occlusion0.02<0.010.020.800.640.2033.31.033.5
    Seq 3:Overlapping targets0.01<0.0101.360.241.2033.11.232.9
    Seq 4:Bad conditions0<0.010.012.642.242.1333.21.132.8
    Seq 5:Multiple objects0<0.0103.152.563.1234.11.133.2
    Total test dataset0.01<0.01<0.011.631.371.3633.51.133.1
    Table 3. Calculation of evaluation index values in different sequence scenarios
    DetectorFAR/%MDR/%FPS
    MPCM38.436.21.45
    RLCM4.6046.21.44
    DNGM3.8945.32.46
    STLCF3.7619.41.52
    Ours<0.011.3633.1
    Table 4. Comparison of different algorithm(without deep learning)
    DetectorFAR/%MDR/%FPS
    YOLOv30.011.6333.5
    RefineDet0.011.561.44
    ISTNet<0.011.3532.4
    FCR-G<0.011.371.11
    Ours<0.011.3633.1
    Table 5. Comparison of different algorithm(with deep learning)
    Zijian ZHU, Qi LIU, Hongfen CHEN, Guiyang ZHANG, Fukuan WANG, Ju HUO. Infrared Small Vehicle Detection Based on Parallel Fusion Network[J]. Acta Photonica Sinica, 2022, 51(2): 0210001
    Download Citation