• Acta Photonica Sinica
  • Vol. 50, Issue 11, 1110001 (2021)
Ming ZHAO1,2 and Haoran ZHANG1,*
Author Affiliations
  • 1College of Information Engineering,Shanghai Maritime University,Shanghai 201306,China
  • 2Key Laboratory of Intelligent Infrared Perception,Shanghai Institute of Technical Physics,Chinese Academy of Sciences,Shanghai 200083,China
  • show less
    DOI: 10.3788/gzxb20215011.1110001 Cite this Article
    Ming ZHAO, Haoran ZHANG. An Infrared Object Detection Method Based on Cross-domain Fusion Network[J]. Acta Photonica Sinica, 2021, 50(11): 1110001 Copy Citation Text show less
    Overall network architecture
    Fig. 1. Overall network architecture
    Comparative learning framework
    Fig. 2. Comparative learning framework
    Soft weight allocation module
    Fig. 3. Soft weight allocation module
    FLIR and Pseudo-visible FLIR
    Fig. 4. FLIR and Pseudo-visible FLIR
    Detection results of FLIR and Pseudo-visible FLIR dataset
    Fig. 5. Detection results of FLIR and Pseudo-visible FLIR dataset
    Comparison of PR curves of various networks
    Fig. 6. Comparison of PR curves of various networks
    Comparison of detection effect on FLIR dataset
    Fig. 7. Comparison of detection effect on FLIR dataset
    Comparison of detection effect on SODA dataset
    Fig. 8. Comparison of detection effect on SODA dataset
    Modal transfer networkCross-domain fusion networksSoft weight allocation modulemAP/%
    ×××78
    ×80.3
    ×81.8
    86.4
    Table 1. Ablation study
    MethodsBackboneDatasetsPrecisionRecallF1-ScoremAP/%
    Mask R-CNNResnet101-FPNFLIR0.6400.9020.74884.0
    Pseudo Visible FLIR0.6170.9210.73883.3
    Table 2. Performance comparison of same network under different datasets
    MethodsBackboneDatasetsAP/%mAP/%FPS
    CarPersonBicycle
    Faster R-CNNResnet101-FPNFLIR83.778.374.778.917
    SODA79.378.573.477.1
    YOLO-V4CSPDarknet-53FLIR8078.875.778.139
    SODA82.778.570.177.1
    MMTODResnet101-FPNFLIR86.883.179.18310
    SODA84.281.47680.5
    BaselineResnet101-FPNFLIR88.384.679.684.214
    SODA84.883.676.581.6
    OursResnet101-FPNFLIR91.388.583.287.716
    SODA8886.980.485.1
    Table 3. Performance comparison of different networks under different datasets
    Ming ZHAO, Haoran ZHANG. An Infrared Object Detection Method Based on Cross-domain Fusion Network[J]. Acta Photonica Sinica, 2021, 50(11): 1110001
    Download Citation