• Laser & Optoelectronics Progress
  • Vol. 58, Issue 4, 0404001 (2021)
Ruihong Guo*, Li Zhang, Ying Yang, Yang Cao, and Junxi Meng
Author Affiliations
  • College of Electronics and Information, Xi'an Polytechnic University, Shaanxi, Xi'an 710048, China
  • show less
    DOI: 10.3788/LOP202158.0404001 Cite this Article Set citation alerts
    Ruihong Guo, Li Zhang, Ying Yang, Yang Cao, Junxi Meng. X-Ray Image Controlled Knife Detection and Recognition Based on Improved SSD[J]. Laser & Optoelectronics Progress, 2021, 58(4): 0404001 Copy Citation Text show less
    Sample photos of the dataset
    Fig. 1. Sample photos of the dataset
    Residual network structure module
    Fig. 2. Residual network structure module
    Feature connection and fusion mode
    Fig. 3. Feature connection and fusion mode
    Improved network model diagram
    Fig. 4. Improved network model diagram
    Jumping connection feature fusion pattern diagram
    Fig. 5. Jumping connection feature fusion pattern diagram
    Experimental results
    Fig. 6. Experimental results
    NetworkSpeed /(frame·s-1)AccuracyMode size /MFLOPS /109
    DenseNet12111474.3321.9
    VGG1618371.753715.3
    MobileNetv223172.0140.52
    ShuffleNet30370.8210.524
    ResNet3440273.1873.6
    Table 1. Results of each network performance test
    HardwareSoftware
    CPU: Intel i7-9750H CPUOperating system
    GTX 1660Ti-8GUbuntu16.04
    GPU: NVIDIA RTX2080tiFrame: Tensorflow
    RAM: 11G×4Language: Python
    Table 2. Experimental environment configuration
    MethodSSDDSSDMFDSSDOur method
    mAP89.891.390.592.6
    Table 3. mAP of each algorithm on SDCK dataset unit: %
    MethodSSDKitchen knifeFruit knifeHacking knifeDaggerScissorSpannerLittle knife
    mAP89.891.088.993.089.287.891.387.5
    Table 4. mAP of SSD algorithm on SDCK controlled tool dataset unit: %
    MethodTrainTestInput sizemAP /%
    SSD2007+20122007300×30078.8
    DSSD2007+20122007321×32180.3
    MFDSSD2007+20122007300×30080.0
    Our method2007+20122007512×51280.5
    Table 5. mAP of each algorithm on VOC2007+2012 dataset
    MethodSSDDSSDMFDSSDOur method
    Speed18.39.613.516.7
    Table 6. Detection speed of each algorithm on SDCK datasetunit: frame·s-1
    Base netFunction modulemAP /%
    VGG1689.8
    ResNet3490.8
    VGG1690.5
    ResNet3492.6
    Table 7. Improved algorithm testmAP step by step
    Ruihong Guo, Li Zhang, Ying Yang, Yang Cao, Junxi Meng. X-Ray Image Controlled Knife Detection and Recognition Based on Improved SSD[J]. Laser & Optoelectronics Progress, 2021, 58(4): 0404001
    Download Citation