• Laser & Optoelectronics Progress
  • Vol. 58, Issue 8, 0810012 (2021)
Yuxiao Wang and Liang Zhang*
Author Affiliations
  • Tianjin Key Laboratory of Intelligent Signal and Image Processing, Civil Aviation University of China, Tianjin 300300, China
  • show less
    DOI: 10.3788/LOP202158.0810012 Cite this Article Set citation alerts
    Yuxiao Wang, Liang Zhang. Dangerous Goods Detection Based on Multi-Scale Feature Fusion in Security Images[J]. Laser & Optoelectronics Progress, 2021, 58(8): 0810012 Copy Citation Text show less
    Structure of SSD model
    Fig. 1. Structure of SSD model
    Structure of MFFNet model
    Fig. 2. Structure of MFFNet model
    Process of fusion module
    Fig. 3. Process of fusion module
    Some pictures in SIXray_OD dataset
    Fig. 4. Some pictures in SIXray_OD dataset
    Training loss function curves of network
    Fig. 5. Training loss function curves of network
    Visual detection results of different models. (a) Original images; (b) SSD model; (c) MFFNet model
    Fig. 6. Visual detection results of different models. (a) Original images; (b) SSD model; (c) MFFNet model
    TypeNumber of images
    GunKnifeWrenchPlierScissorTotal
    Training20551092158627658116102
    Test88146968011863482616
    Total293615612266395111598718
    Table 1. Number of images of different types in datasets
    CombinationBased layerExtra layer
    Block 30Block 33Conv8Conv9Conv10
    1
    2
    3
    Table 2. Combination of different fusion feature layers
    CombinationSumProductConcat
    178.2777.7678.10
    277.8477.5377.92
    378.0577.0277.54
    Table 3. Detection accuracy results of different fusion methods unit: %
    TypeSSDMFFNet
    Gun89.9190.42
    Knife73.3175.29
    Wrench69.4671.17
    Plier75.5782.22
    Scissor63.6172.24
    mAP74.3778.27
    Table 4. Detection accuracy results of all kinds of contraband unit: %
    ModelBackbonemAP/%
    SSDVGG-1674.37
    SSDResNet-10176.18
    SSD+FM 1ResNet-10177.38
    SSD+FM 2ResNet-10177.75
    SSD+FM 1+FM 2ResNet-10178.27
    Table 5. Results of ablation experiment
    ModelmAP /%FPS
    SSD74.3756
    FSSD75.7541
    Faster R-CNN77.812
    YOLO-v370.4970
    MFFNet78.2719
    Table 6. Detection results of different models
    Yuxiao Wang, Liang Zhang. Dangerous Goods Detection Based on Multi-Scale Feature Fusion in Security Images[J]. Laser & Optoelectronics Progress, 2021, 58(8): 0810012
    Download Citation