• Laser & Optoelectronics Progress
  • Vol. 57, Issue 10, 101010 (2020)
Xun Li1, Binbin Shi1、*, Yang Liu2, Lei Zhang1, and Xiaohua Wang1
Author Affiliations
  • 1School of Electronics and Information, Xi'an Polytechnic University, Xi'an, Shaanxi 710048, China
  • 2Xi'an Metrological Technology Research Institute, Xi'an, Shaanxi 710068, China
  • show less
    DOI: 10.3788/LOP57.101010 Cite this Article Set citation alerts
    Xun Li, Binbin Shi, Yang Liu, Lei Zhang, Xiaohua Wang. Multi-Target Recognition Method Based on Improved YOLOv2 Model[J]. Laser & Optoelectronics Progress, 2020, 57(10): 101010 Copy Citation Text show less
    Target position
    Fig. 1. Target position
    Test results of different models. (a) Model 1; (b) model 2; (c) model 3; (d) model 4; (e) model 5; (f) model 6
    Fig. 2. Test results of different models. (a) Model 1; (b) model 2; (c) model 3; (d) model 4; (e) model 5; (f) model 6
    Frame of YOLOv2-voc_mul model
    Fig. 3. Frame of YOLOv2-voc_mul model
    Model data sample. (a) Car; (b) van; (c) bus; (d) truck
    Fig. 4. Model data sample. (a) Car; (b) van; (c) bus; (d) truck
    Loss graphs of different models. (a) YOLOv2; (b) YOLOv2-voc; (c) YOLOv3; (d) YOLOv2-voc_mul
    Fig. 5. Loss graphs of different models. (a) YOLOv2; (b) YOLOv2-voc; (c) YOLOv3; (d) YOLOv2-voc_mul
    Verification results of different models. (a) YOLOv2; (b) YOLOv2-voc; (c) YOLOv3; (d) YOLOv2-voc_mul
    Fig. 6. Verification results of different models. (a) YOLOv2; (b) YOLOv2-voc; (c) YOLOv3; (d) YOLOv2-voc_mul
    Test results at different number of iterations. (a) 60000; (b) 70000
    Fig. 7. Test results at different number of iterations. (a) 60000; (b) 70000
    Added model samples
    Fig. 8. Added model samples
    Test results of different models. (a) YOLOv2 model; (b) YOLOv2-voc model; (c) YOLOv3 model; (d) YOLOv2-voc_mul model
    Fig. 9. Test results of different models. (a) YOLOv2 model; (b) YOLOv2-voc model; (c) YOLOv3 model; (d) YOLOv2-voc_mul model
    ModelLearning rateLayer number of network structureActivation function
    ConvolutionlayerMaximum pooling layer+average pooling layerBN layer
    Model 10.001235+02222 leaky+1 linear
    Model 20.0001235+02022 leaky+1 linear
    Model 30.01235+02222 leaky+1 linear
    Model 40.001205+12219 leaky+1 linear
    Model 50.001205+12219 leaky+1 ReLU
    Model 60.001205+02019 leaky+1 linear
    Table 1. Network framework
    ModelNtotalCcorrectPproposalPprecision /%Rrecall /%F1 /%
    YOLOv215214615196.6996.0596.36
    YOLOv2-voc15214414698.6394.7496.64
    YOLOv31528515056.6755.9269.68
    YOLOv2-voc_mul15214514699.2095.3997.26
    Table 2. Test results
    ModelmAP /%
    TruckBusVanCar
    YOLOv286.4584.1682.3786.88
    YOLOv2-voc87.6185.2283.0287.31
    YOLOv383.3581.8377.9683.92
    YOLOv2-voc_mul88.7288.5686.6489.03
    Table 4. Comparison of mAP value in different models
    TypeAccuracy /%Averageaccuracy /%
    TruckBusVanCar
    Simple target92.1491.8990.0894.7292.21
    Multiple target89.0188.7688.2091.7189.44
    Table 5. Average accuracy of vehicle identification in the YOLOv2-voc_mul model
    Xun Li, Binbin Shi, Yang Liu, Lei Zhang, Xiaohua Wang. Multi-Target Recognition Method Based on Improved YOLOv2 Model[J]. Laser & Optoelectronics Progress, 2020, 57(10): 101010
    Download Citation