• Acta Photonica Sinica
  • Vol. 50, Issue 1, 173 (2021)
Shuigen WEI1, Chengwei WANG1、*, Zhen CHEN1, Congxuan ZHANG1, and Xiaoyu ZHANG2
Author Affiliations
  • 1Key Laboratory of Nondestructive Testing, Ministry of Education, Nanchang Hangkong University, Nanchang330063, China
  • 2School of Information Engineering, Southwest University of Science and Technology, Mianyang, Sichuan61010, China
  • show less
    DOI: 10.3788/gzxb20215001.0110001 Cite this Article
    Shuigen WEI, Chengwei WANG, Zhen CHEN, Congxuan ZHANG, Xiaoyu ZHANG. Infrared Dim Target Detection Based on Human Visual Mechanism[J]. Acta Photonica Sinica, 2021, 50(1): 173 Copy Citation Text show less
    Some images of MSIDT
    Fig. 1. Some images of MSIDT
    Statistical properties of MSIDT
    Fig. 2. Statistical properties of MSIDT
    Source image and Saliency map
    Fig. 3. Source image and Saliency map
    Detection results before NMS
    Fig. 4. Detection results before NMS
    Detection results after NMS
    Fig. 5. Detection results after NMS
    ROI divided into cells
    Fig. 6. ROI divided into cells
    Soft-FART network
    Fig. 7. Soft-FART network
    Algorithm flow chart
    Fig. 8. Algorithm flow chart
    Part of training samples
    Fig. 9. Part of training samples
    Influence of parameters t and Ra on detection performance in different background
    Fig. 10. Influence of parameters t and Ra on detection performance in different background
    Influence of parameters ρ on detection performance in different background
    Fig. 11. Influence of parameters ρ on detection performance in different background
    Detection results of different methods
    Fig. 12. Detection results of different methods
    MethodArchitecture

    Cloudless

    sky

    Complex

    cloud

    Continuous

    cloud_sky

    Trees

    Others

    (sea/sea_sky)

    Average
    FART0.8380.990.9020.8790.7610.7440.852
    SVM0.8320.990.7860.9200.7290.6890.824
    ANN0.8201.0000.8150.8770.7510.6450.818
    Soft-FART0.9281.0000.9350.9470.9310.8580.933
    Table 1. Comparison of F1-score between Soft-FART and FART
    OursLCMILCMMPCMLIGHB-MLCM
    Backgroundpcpcpcpcpcpc
    Architecture0.940.910.440.550.450.500.330.760.260.500.280.19
    Cloudless_sky1.001.000.980.631.000.990.971.001.000.990.970.99
    Complex_cloud0.970.900.250.820.980.670.490.650.930.800.830.84
    Continuous_cloud sky0.970.930.380.500.920.770.470.860.880.840.770.78
    Trees0.980.880.250.320.420.610.450.540.250.520.290.36
    Others(sea/sea_sky)0.990.760.320.510.900.800.790.620.930.650.870.55
    Average0.970.890.440.550.780.720.580.740.710.720.670.62
    Table 2. Precision(p) and call(c) of different methods (%)
    BackgroundOursLCMILCMMPCMLIGHB-MLCM
    Architecture0.9280.4880.4720.4620.3410.216
    Cloudless_sky1.0000.7670.9980.9840.9950.980
    Complex_cloud0.9350.38190.7940.5590.8600.839
    Continuous_cloud_sky0.9470.4290.8370.6050.8570.777
    Trees0.9310.2800.4950.4910.3390.318
    Others(sea/sea_sky)0.8580.3960.8480.6930.7630.672
    Average0.9330.4570.7410.6320.6930.634
    Table 3. F1-score of different methods
    BackgroundOursLCMILCMMPCMLIGHB-MLCM
    Architecture45.32720.7153.44161.601314.7627.86
    Cloudless_sky35.82691.3149.10154.061281.6227.18
    Complex_cloud35.67651.1047.47144.231195.1725.73
    Continuous_cloud_sky36.93683.5550.18156.851266.6025.66
    Trees46.28743.1754.29171.121365.0629.64
    Others(sea/sea_sky)26.78403.1730.54114.85482.6417.95
    Average37.80648.8347.50150.451150.9825.67
    Table 4. Detection efficiency of different methods (ms)
    Shuigen WEI, Chengwei WANG, Zhen CHEN, Congxuan ZHANG, Xiaoyu ZHANG. Infrared Dim Target Detection Based on Human Visual Mechanism[J]. Acta Photonica Sinica, 2021, 50(1): 173
    Download Citation