Laser & Optoelectronics Progress, Volume. 60, Issue 6, 0630001(2023)

Application of Improved Auto-Encoding Network Feature Extraction Method in Near Infrared Spectral Quantitative Analysis

Zhiyong Luo1... Yuhua Qin1,*, Shijie Wang1, Susu He1 and Haitao Zhang2 |Show fewer author(s)
Author Affiliations
  • 1College of Information Science and Technology, Qingdao University of Science & Technology, Qingdao 266061, Shandong, China
  • 2Technical Research Center, China Tobacco Yunnan Industrial Co., Ltd., Kunming 650024, Yunnan, China
  • show less
    Figures & Tables(11)
    Auto-encoding network model
    Convolutional auto-encoding network
    1D-BCAE structure
    Flow chart of 1D-BCAE
    Original near infrared spectra of cigarette samples
    Pretreatment spectra of SG+1st derivative
    Predicted value-true value curve. (a) PCA; (b) CAE; (c) 1D-BCAE
    • Table 1. Comparison results of nicotine pretreatment methods

      View table

      Table 1. Comparison results of nicotine pretreatment methods

      Pretreatment methodRoot mean square error of cross validation
      SNV+1st derivative0.1357
      SNV+2nd derivative0.1384
      SG+1st derivative0.0987
      SG+2nd derivative0.1172
      MSC+1st derivative0.1401
      MSC+2nd derivative0.1524
      No pretreatment0.3453
    • Table 2. Reconstruction error and root mean square error of different models

      View table

      Table 2. Reconstruction error and root mean square error of different models

      NetworkReconstruction error /%Root mean square error
      AE47.530.042
      CAE35.790.027
      1D-BCAE17.260.016
    • Table 3. Comparison of training duration

      View table

      Table 3. Comparison of training duration

      ModelNumber of iterationsTotal time /s
      Model with BN5001215
      Model without BN10002186
    • Table 4. Comparison of quantitative modeling performance of different feature subsets

      View table

      Table 4. Comparison of quantitative modeling performance of different feature subsets

      MethodNicotineTotal sugar
      RMSECVRMSEPR2Mean prediction error /%RMSECVRMSEPR2Mean prediction error /%
      Full spectrum0.30030.31540.95236.5230.71350.72150.95126.344
      PCA0.24350.25630.96015.8780.66580.63680.96015.022
      CAE0.14010.14210.97134.5020.49670.47820.97124.124
      1D-BCAE0.09870.11250.98793.7130.39470.38920.98912.876
    Tools

    Get Citation

    Copy Citation Text

    Zhiyong Luo, Yuhua Qin, Shijie Wang, Susu He, Haitao Zhang. Application of Improved Auto-Encoding Network Feature Extraction Method in Near Infrared Spectral Quantitative Analysis[J]. Laser & Optoelectronics Progress, 2023, 60(6): 0630001

    Download Citation

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

    Category: Spectroscopy

    Received: Feb. 15, 2022

    Accepted: Mar. 29, 2022

    Published Online: Mar. 7, 2023

    The Author Email: Yuhua Qin (yuu71@163.com)

    DOI:10.3788/LOP220740

    Topics