• Laser & Optoelectronics Progress
  • Vol. 58, Issue 8, 0810003 (2021)
Shouxiang Guo and Liang Zhang*
Author Affiliations
  • Tianjin Key Laboratory of Advanced Signal & Image Processing, Civil Aviation University of China, Tianjin 300300, China
  • show less
    DOI: 10.3788/LOP202158.0810003 Cite this Article Set citation alerts
    Shouxiang Guo, Liang Zhang. Yolo-C: One-Stage Network for Prohibited Items Detection Within X-Ray Images[J]. Laser & Optoelectronics Progress, 2021, 58(8): 0810003 Copy Citation Text show less
    Yolov3 network structure
    Fig. 1. Yolov3 network structure
    Composite backbone network structure
    Fig. 2. Composite backbone network structure
    Feature augment block
    Fig. 3. Feature augment block
    Yolo-C network structure of one-stage dual-network object detection algorithm
    Fig. 4. Yolo-C network structure of one-stage dual-network object detection algorithm
    Random X-ray images of GDXray dataset
    Fig. 5. Random X-ray images of GDXray dataset
    Random X-ray images of SIXray dataset
    Fig. 6. Random X-ray images of SIXray dataset
    Sample display of prohibited items
    Fig. 7. Sample display of prohibited items
    Comparison graph of network training process
    Fig. 8. Comparison graph of network training process
    IoU definition and calculation diagram
    Fig. 9. IoU definition and calculation diagram
    Detection results of experiments 3, 5 and 7
    Fig. 10. Detection results of experiments 3, 5 and 7
    TypeLayerFilter numberSizeOutput
    DBLConv323×3416×416
    Res1Conv2643×3, 1×1208×208
    Upsample323×3416×416
    Res1'Conv2643×3, 1×1208×208
    Res2Conv41283×3, 1×1104×104
    Upsample643×3208×208
    Res2'Conv41283×3, 1×1104×104
    Res8Conv162563×3, 1×152×52
    Upsample1283×3104×104
    Res8'Conv162563×3, 1×152×52
    Res8Conv165123×3, 1×126×26
    Upsample2563×352×52
    Res8'Conv165123×3, 1×126×26
    Res4Conv810243×3, 1×113×13
    Upsample5123×326×26
    Res4'Conv810243×3, 1×113×13
    DBLConv303×313×13
    Head
    FABUpsample3×3, 1×126×26
    DBLConv303×326×26
    Head
    FABUpsample3×3, 1×152×52
    DBLConv303×352×52
    Head
    Table 1. Parameters of Yolo-C
    ModelBackboneFABFLOPsParams/106
    Yolov3DarkNet-5332.7761.55
    OursDarkNet-5335.5864.65
    DarkNet-C61.85105.93
    Table 2. Analysis of different network complexity
    No.ModelBackboneFABAP for gun /%AP for knife /%AP for pliers /%AP for wrench /%AP for scissor /%mAP /%Detection rate /(frame·s-1)
    1SSDResNet5085.6972.4451.2160.6241.9962.3956
    2FASFResNet10182.7770.2348.657.738.759.648
    3Yolov3DarkNet-5388.6776.5352.4961.4142.664.3457
    4Faster-RCNNResNet10193.8383.7458.8571.5854.172.1810
    5OursDarkNet-5390.179.5755.6764.8151.0068.2355
    6DarkNet-C91.682.158.769.754.471.1042
    7DarkNet-C93.1483.1260.1873.8258.1473.6840
    Table 3. Ablation experiments on the Yolo-C network based on SIXray_OD dataset
    Shouxiang Guo, Liang Zhang. Yolo-C: One-Stage Network for Prohibited Items Detection Within X-Ray Images[J]. Laser & Optoelectronics Progress, 2021, 58(8): 0810003
    Download Citation