• Acta Photonica Sinica
  • Vol. 49, Issue 7, 710003 (2020)
Jun-ming LIU1 and Wei-hua MENG1、2
Author Affiliations
  • 1China Airborne Missile Academy, Luoyang, Henan 471009, China
  • 2Aviation Key Laboratory of Science and Technology on Airborne Guided Weapons, Luoyang, Henan 471009, China
  • show less
    DOI: 10.3788/gzxb20204907.0710003 Cite this Article
    Jun-ming LIU, Wei-hua MENG. Infrared Small Target Detection Based on Fully Convolutional Neural Network and Visual Saliency[J]. Acta Photonica Sinica, 2020, 49(7): 710003 Copy Citation Text show less
    Proposed detection scheme
    Fig. 1. Proposed detection scheme
    Detection schemes of different algorithms
    Fig. 2. Detection schemes of different algorithms
    Typical fully convolutional neural networks for infrared small target detection.
    Fig. 3. Typical fully convolutional neural networks for infrared small target detection.
    Proposed fully convolutional neural network
    Fig. 4. Proposed fully convolutional neural network
    Typical target image based on 2D Gaussian model
    Fig. 5. Typical target image based on 2D Gaussian model
    Typical infrared images with small targets
    Fig. 6. Typical infrared images with small targets
    Segmentation results of different networks
    Fig. 7. Segmentation results of different networks
    ROC curve of proposed algorithms with and without contrast feature
    Fig. 8. ROC curve of proposed algorithms with and without contrast feature
    Test images and results of seven algorithms
    Fig. 9. Test images and results of seven algorithms
    ROC curves of different algorithms
    Fig. 10. ROC curves of different algorithms
    TypeDescriptionPdFaGFLOPs
    M1Base network in Fig. 40.940 87.97×10-60.235
    M11Without down-sampling layer0.938 92.01×10-50.494
    M12Without feature of 3rd conv layer0.929 41.11×10-50.226
    M13Without SE layer0.938 91.42×10-50.235
    Table 1. Comparation of different design choices of proposed network
    TypeDescriptionPdFaGFLOPs
    M1Network in Fig. 40.940 87.97×10-60.235
    M2Network in Fig. 3(a)0.931 31.06×10-56.47
    M3Network in Fig. 3(b)0.933 24.23×10-51.54
    Table 2. Performance of different networks
    IndexADMDIPIMLIGMaxMedianFMDNNFCN+CNNS-FCN
    120.378.931.012.025.163.6118.0
    217.900.14.815.221.028.9
    310.138.96.412.75.317.155.9
    416.917.920.515.919.025.925.3
    58.212.59.61.85.96.912.6
    656.10.01.950.333.5310.8520.7
    75.224.23.819.57.29.929.8
    812.537.035.214.54.419.751.3
    Table 3. G of different algorithms on 8 test images
    IndexADMDIPIMLIGMaxMedianFMDNNFCN+CNNS-FCN
    14.219.96.32.55.112.823.7
    25.3--1.44.56.28.5
    34.119.65.05.12.16.822.3
    46.18.27.45.86.99.49.2
    52.23.42.60.91.61.93.4
    63.7--3.01.917.929.3
    72.313.63.18.33.14.312.7
    86.627.718.47.62.410.326.8
    Table 4. B of different algorithms on 8 test images
    MethodADMDIPIMLIGMaxMedianFMDNNFCN+CNNS-FCN
    Time/s0.9960.8660.2840.2190.0660.1640.033
    Table 5. Average run time of different algorithms
    Jun-ming LIU, Wei-hua MENG. Infrared Small Target Detection Based on Fully Convolutional Neural Network and Visual Saliency[J]. Acta Photonica Sinica, 2020, 49(7): 710003
    Download Citation