Dangerous Goods Detection Based on Multi-Scale Feature Fusion in Security Images

Table 1. Number of images of different types in datasets
View table
Table 1. Number of images of different types in datasets
Type Number of images
Gun Knife Wrench Plier Scissor Total
Training 2055 1092 1586 2765 811 6102
Test 881 469 680 1186 348 2616
Total 2936 1561 2266 3951 1159 8718

Table 2. Combination of different fusion feature layers
View table
Table 2. Combination of different fusion feature layers
Combination Based layer Extra layer
Block 30 Block 33 Conv8 Conv9 Conv10
1 √ √
√ √
2 √ √
√ √
3 √ √
√ √

Table 3. Detection accuracy results of different fusion methods unit: %
View table
Table 3. Detection accuracy results of different fusion methods unit: %
Combination Sum Product Concat
1 78.27 77.76 78.10
2 77.84 77.53 77.92
3 78.05 77.02 77.54

Table 4. Detection accuracy results of all kinds of contraband unit: %
View table
Table 4. Detection accuracy results of all kinds of contraband unit: %
Type SSD MFFNet
Gun 89.91 90.42
Knife 73.31 75.29
Wrench 69.46 71.17
Plier 75.57 82.22
Scissor 63.61 72.24
mAP 74.37 78.27

Table 5. Results of ablation experiment
View table
Table 5. Results of ablation experiment
Model Backbone mAP/%
SSD VGG-16 74.37
SSD ResNet-101 76.18
SSD+FM 1 ResNet-101 77.38
SSD+FM 2 ResNet-101 77.75
SSD+FM 1+FM 2 ResNet-101 78.27

Table 6. Detection results of different models
View table
Table 6. Detection results of different models
Model mAP /% FPS
SSD 74.37 56
FSSD 75.75 41
Faster R-CNN 77.81 2
YOLO-v3 70.49 70
MFFNet 78.27 19

Tools

Get Citation

Copy Citation Text

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

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites