• Laser & Optoelectronics Progress
  • Vol. 59, Issue 2, 0210011 (2022)
Wen Wang1, Yatong Zhou1、*, Baojun Shi2, Hao He1, and Jianwei Zhang1
Author Affiliations
  • 1School of Electronic Information Engineering, Hebei University of Technology, Tianjin 300401, China
  • 2School of Mechanical Engineering, Hebei University of Technology, Tianjin 300401, China
  • show less
    DOI: 10.3788/LOP202259.0210011 Cite this Article Set citation alerts
    Wen Wang, Yatong Zhou, Baojun Shi, Hao He, Jianwei Zhang. Recognition Algorithm of Dangerous Goods in Security Inspection Based on Multi-Layer Attention Mechanism[J]. Laser & Optoelectronics Progress, 2022, 59(2): 0210011 Copy Citation Text show less
    Structure of the dangerous goods security inspection algorithm based on multi-layer attention mechanism
    Fig. 1. Structure of the dangerous goods security inspection algorithm based on multi-layer attention mechanism
    Principle of the channel attention mechanism
    Fig. 2. Principle of the channel attention mechanism
    Principle of the spatial attention mechanism
    Fig. 3. Principle of the spatial attention mechanism
    Structure of the ResNet101
    Fig. 4. Structure of the ResNet101
    Images in the security image data set
    Fig. 5. Images in the security image data set
    Number of five types of dangerous goods
    Fig. 6. Number of five types of dangerous goods
    Loss values in the training process of six algorithms
    Fig. 7. Loss values in the training process of six algorithms
    Optimal accuracy of six algorithms
    Fig. 8. Optimal accuracy of six algorithms
    Visualization result of feature map. (a) Example 1; (b) example 2
    Fig. 9. Visualization result of feature map. (a) Example 1; (b) example 2
    AlgorithmResNet101CHRA1A2A3mAP/%
    ResNet101+CHR81.57
    ResNet101+CHR+A181.46
    ResNet101+CHR+A282.99
    ResNet101+CHR+A382.97
    ResNet101+CHR+A2A383.26
    ResNet101+CHR+A1A2A382.79
    Table 1. Results of ablation experiments
    AlgorithmmAP

    SIFT+SVM

    ResNet50+CHR

    73.65

    80.68

    ResNet101+CHR

    DenseNet+CHR

    ResNet50+CHR+A2A3(ours)

    ResNet101+CHR+A2A3(ours)

    81.57

    82.06

    82.34

    83.26

    DenseNet+CHR+A2A3(ours)83.89
    Table 2. Results of comparative experiment
    Wen Wang, Yatong Zhou, Baojun Shi, Hao He, Jianwei Zhang. Recognition Algorithm of Dangerous Goods in Security Inspection Based on Multi-Layer Attention Mechanism[J]. Laser & Optoelectronics Progress, 2022, 59(2): 0210011
    Download Citation