Journals Articles Conferences News About CLP
Submit a paper Email Alert
Search
Search
Articles
Journals
News
Advanced Search
Top Searches
2D MaterialsTransformation opticsQuantum PhotonicsSmall lasersSemiconductor UV PhotonicsSupersymmetric microring laser arrays

Laser & Optoelectronics Progress, Volume. 55, Issue 12, 121009(2018)

Traffic Sign Recognition Based on Improved Deep Convolution Neural Network

Yongjie Ma*, Xueyan Li, and Xiaofeng Song
Author Affiliations
  • College of Physics and Electronic Engineering, Northwest Normal University, Lanzhou, Gansu 730070, China
    • show less
    AbstractGet PDF(in Chinese)
    Figures&Tables (14)
    Equations (0)
    References (17)
    Cited By (7)
    Tools
    Paper Information
    Topics
    Figures & Tables(14)
    Maximum pooling schematic

    Fig. 1. Maximum pooling schematic

    Download full size
    Basic structure of AlexNet network

    Fig. 2. Basic structure of AlexNet network

    Download full size
    Neural network contrast diagrams. Three-level neural network with (a) unused and (b) used dropout

    Fig. 3. Neural network contrast diagrams. Three-level neural network with (a) unused and (b) used dropout

    Download full size
    Structure diagram of improved AlexNet model

    Fig. 4. Structure diagram of improved AlexNet model

    Download full size
    Experimental flow chart

    Fig. 5. Experimental flow chart

    Download full size
    Visualization of the coiling layer operation

    Fig. 6. Visualization of the coiling layer operation

    Download full size
    Contrast diagrams of (a) Accuracy and (b) loss

    Fig. 7. Contrast diagrams of (a) Accuracy and (b) loss

    Download full size

    • Table 1. Model setting of improved AlexNet network

      View table

      Table 1. Model setting of improved AlexNet network

      LayerLayerinputConvolution kernelConvolutionoutputPoolingPooledoutputLayeroutput
      SizeNumberStepPadSizeMode
      L1(conv1+pool1)48×48×33×3841148×48×842×2Max24×24×8424×24×84
      L2(conv2+pool2)24×24×843×31251124×24×1252×2Max12×12×12512×12×125
      L3(conv3)12×12×1253×32501010×10×250———10×10×250
      L4(conv4)10×10×2503×35001110×10×500———10×10×500
      L5(conv5+pool5)10×10×5003×3250108×8×2502×2Max4×4×2504×4×250
      L6(conv6)4×4×2503×3250102×2×250———2×2×250
      L7(conv7)2×2×2502×2500101×1×500———1×1×500
      L8(Full)1×1×500————————1000
      L9(Full)1000————————1000
      L10(Softmax)1000————————44

    • Table 2. Impact of dropout on the model

      View table

      Table 2. Impact of dropout on the model

      ModelAlexNet modelImproved model without dropoutImproved model with dropout
      Test sample error number19617358
      Test error rate0.0400.0350.012

    • Table 3. Algorithm classification ability analysis

      View table

      Table 3. Algorithm classification ability analysis

      AlgorithmTrainingtime /hParameter consumptionmemory /MBIdentifying eachimage time /msRecognitionaccuracy /%
      AlexNet model173228.215895.568
      Improved model16214096.875

    • Table 4. Algorithm comparison and analysis

      View table

      Table 4. Algorithm comparison and analysis

      AlgorithmTraining time /hParameter consumption memory /MBRecognition accuracy /%
      Improved model162196.875
      Contrast model 22121.7—
      Contrast model 1—21.293.750

    • Table 5. Comparison of different methods in GTSRB dataset recognition results

      View table

      Table 5. Comparison of different methods in GTSRB dataset recognition results

      AlgorithmClassification time /msAccuracy rate /%
      HOG+SVM algorithm of Ref. [17]17695.68
      ANN—89.63
      Random forests—96.14
      Improved model algorithm4096.875
      Algorithm of Ref. [1]27599.01
      Algorithm of Ref. [6]15298.57

    • Table 6. Identification of AlexNet model under real environmental conditions

      View table

      Table 6. Identification of AlexNet model under real environmental conditions

      TypeTest sample numberCorrect recognition number ofAlexNet modelRecognition accuracyrate of AlexNet model /%
      Motion blur302273.3
      Background interference514384.3
      Light (weather)302583.3
      Shooting angle302686.6

    • Table 7. Identification of improved model under real environmental conditions

      View table

      Table 7. Identification of improved model under real environmental conditions

      TypeTest sample numberCorrect recognition number ofimproved modelRecognition accuracy rate ofimproved model /%
      Motion blur302376.6
      Background interference514588.2
      Light (weather)302790.0
      Shooting angle302893.3
    Tools

    Get Citation

    Copy Citation Text

    Yongjie Ma, Xueyan Li, Xiaofeng Song. Traffic Sign Recognition Based on Improved Deep Convolution Neural Network[J]. Laser & Optoelectronics Progress, 2018, 55(12): 121009

    Download Citation

    EndNote(RIS)BibTexPlain Text
    Set citation alerts for article
    Save article for my favorites
    Paper Information

    Category: Image Processing

    Received: Apr. 25, 2018

    Accepted: Jul. 12, 2018

    Published Online: Aug. 1, 2019

    The Author Email: Yongjie Ma (myjmyj@163.com)

    DOI:10.3788/LOP55.121009

    Topics

    laser devices and laser physics
    Lasers and Laser Optics
    Laser physics
    laser manufacturing
    Instrumentation, Measurement and Metrology