• Laser & Optoelectronics Progress
  • Vol. 58, Issue 16, 1610012 (2021)
Caizhen Zhang, Ying Li*, binlong Kang, and yuan Chang
Author Affiliations
  • School of Electronics and Information Engineering, Lanzhou Jiaotong University, Lanzhou, Gansu 730073, China
  • show less
    DOI: 10.3788/LOP202158.1610012 Cite this Article Set citation alerts
    Caizhen Zhang, Ying Li, binlong Kang, yuan Chang. Blurred License Plate Character Recognition Algorithm Based on Deep Learning[J]. Laser & Optoelectronics Progress, 2021, 58(16): 1610012 Copy Citation Text show less
    CNN configuration diagram. (a) Depthwise separable convolution bolck; (b) CNN structure
    Fig. 1. CNN configuration diagram. (a) Depthwise separable convolution bolck; (b) CNN structure
    Structure of Bi-LSTM network model
    Fig. 2. Structure of Bi-LSTM network model
    Process of CRNN algorithm
    Fig. 3. Process of CRNN algorithm
    Sample dataset example. (a) Samples of GAN_LP ; (b) samples of Reldv1; (c) samples of CCPDv1
    Fig. 4. Sample dataset example. (a) Samples of GAN_LP ; (b) samples of Reldv1; (c) samples of CCPDv1
    DatesetTraining setValidation setTest set
    GAN-LP122009807406
    Reldv125000500
    CCPDv1300001000
    Table 1. Statistics of datasets
    ParameterSetting
    Base_lr0.001
    γ0.1
    Lr_policyMultistep
    Active_funReLU
    Iterations2×104
    TypeSGD
    Weight_decay0.0005
    Momentum0.8
    BNYes
    Batch_size64
    Table 2. Hyper parameters setting
    DatasetGAN-LPCCPDv1Reldv1
    ACRR97.697.898.2
    CCRR95.193.395.7
    fRA96.896.997.8
    fCRA97.197.298.4
    Table 3. Comprehensive experimental results of improved CRNN+CTC network unit: %
    MethodGAN_LPCCPDv1Reldv1
    RA%CRA%T /msRA%CRA%T /msRA%CRA%T /ms
    Traditional[10]84.277.32583.474.23185.575.732
    LPR-Net[15]94.392.76287.486.67288.586.346
    CRNN+CTC[18]96.996.24697.196.34896.497.044
    Ours96.897.12996.997.23397.898.437
    Table 4. Experimental comparison results
    Caizhen Zhang, Ying Li, binlong Kang, yuan Chang. Blurred License Plate Character Recognition Algorithm Based on Deep Learning[J]. Laser & Optoelectronics Progress, 2021, 58(16): 1610012
    Download Citation