• Spectroscopy and Spectral Analysis
  • Vol. 41, Issue 7, 2188 (2021)
Li-jia XU1、*, Ming CHEN1、1;, Yu-chao WANG1、1;, Xiao-yan CHEN2、2; 3;, and Xiao-long LEI1、1; *;
Author Affiliations
  • 11. College of Mechanical and Electrical Engineering, Sichuan Agricultural University, Ya'an 625014, China
  • 22. College of Information Engineering, Sichuan Agricultural University, Ya'an 625014, China
  • show less
    DOI: 10.3964/j.issn.1000-0593(2021)07-2188-08 Cite this Article
    Li-jia XU, Ming CHEN, Yu-chao WANG, Xiao-yan CHEN, Xiao-long LEI. Study on Non-Destructive Detection Method of Kiwifruit Sugar Content Based on Hyperspectral Imaging Technology[J]. Spectroscopy and Spectral Analysis, 2021, 41(7): 2188 Copy Citation Text show less
    Hyperspectral sorter
    Fig. 1. Hyperspectral sorter
    Spectral information of kiwi samples
    Fig. 2. Spectral information of kiwi samples
    Spectral reflectance images before and after DOSC preprocessing
    Fig. 3. Spectral reflectance images before and after DOSC preprocessing
    The correlation coefficient between spectral band and sugar content before and after DOSC preprocessing
    Fig. 4. The correlation coefficient between spectral band and sugar content before and after DOSC preprocessing
    Extracting characteristic variables by IRIV
    Fig. 5. Extracting characteristic variables by IRIV
    Distribution of IRIV characteristic spectral variables
    Fig. 6. Distribution of IRIV characteristic spectral variables
    The results of CARS
    Fig. 7. The results of CARS
    Distribution diagram of SPA spectral characteristic variables
    Fig. 8. Distribution diagram of SPA spectral characteristic variables
    Distribution of SPA characteristic spectral variables for kiwifruit sugar content
    Fig. 9. Distribution of SPA characteristic spectral variables for kiwifruit sugar content
    Data setSamplesMinMaxMeanS.D
    Calibration set909.5018.1013.171.93
    Prediction set3010.4016.5013.561.52
    Table 1. The statistical results of kiwi fruit sugar content measurement (unit: /°Brix)
    Pretreatment
    method
    Number of
    hidden
    neurons
    Calibration
    set
    Prediction
    set
    RMSECRCRMSEPRP
    Raw valid data921.076 20.673 41.126 00.539 9
    SNV721.418 20.410 21.414 90.342 1
    MSC711.345 70.469 01.154 70.561 8
    DOSC350.518 40.927 30.753 20.744 9
    Table 2. All-band ELM prediction model using different pretreatment methods
    Extraction methodVariable number
    CARS49
    SPA9
    IRIV8
    CARS+SPA58
    CARS+IRIV55
    (CARS+SPA)-SPA11
    (CARS+IRIV)-SPA19
    Table 3. The number of feature variables extracted by different feature extraction methods
    Extraction
    method
    Variable
    number
    γσ2
    CARS49178 232.022 62 315.787 8
    SPA924 689.722 84 419.308 9
    IRIV868 345.149 34 095.753 5
    CARS+SPA58128 353.819 6519.357 2
    CARS+IRIV5566 321.544 43 244.478 3
    (CARS+SPA)-SPA11213 051.988 12 541.380 7
    (CARS+IRIV)-SPA19292 698.908 2810.048 4
    Table 4. PSO optimized parameters γ, σ2
    Extraction methodVariable numberRMSECRCRMSEPRPRPD
    CARS490.658 30.882 70.878 10.653 31.727 4
    SPA90.672 60.877 60.881 60.650 61.720 6
    IRIV80.652 50.884 80.858 20.668 81.767 4
    CARS+SPA580.659 50.882 30.886 60.646 61.710 9
    CARS+IRIV550.654 20.884 20.871 60.658 41.740 3
    (CARS+SPA)-SPA110.662 50.881 20.895 90.639 11.693 1
    (CARS+IRIV)-SPA190.666 60.879 70.883 10.649 41.717 7
    Table 5. Prediction results of SVR model established by different feature extraction methods
    Extraction methodVariable numberRMSECRCRMSEPRPRPD
    CARS490.451 50.944 80.834 80.686 71.817 0
    SPA90.637 30.890 10.874 50.656 21.734 5
    IRIV80.624 00.894 60.867 10.662 01.749 3
    CARS+SPA580.138 70.994 80.823 10.695 41.842 8
    CARS+IRIV550.471 90.939 70.824 80.694 11.839 0
    (CARS+SPA)-SPA110.580 50.908 80.870 70.659 11.742 1
    (CARS+IRIV)-SPA190.471 50.939 80.788 50.720 41.923 5
    Table 6. Prediction results of LSSVM model established by different feature extraction methods
    Extraction methodVariable numberNumber of hidden neuronsRMSECRCRMSEPRPRPD
    CARS49460.449 70.945 30.685 00.789 02.214 4
    SPA9890.585 40.907 20.799 30.712 71.897 7
    IRIV8860.557 30.916 00.796 80.714 51.903 6
    CARS+SPA58500.466 60.941 10.637 20.817 42.380 4
    CARS+IRIV55500.449 20.945 40.591 80.842 52.563 0
    (CARS+SPA)-SPA11490.511 80.929 10.688 60.786 82.202 6
    (CARS+IRIV)-SPA19920.450 30.945 10.598 30.839 02.535 1
    Table 7. Prediction results of ELM model established by different feature extraction methods
    Li-jia XU, Ming CHEN, Yu-chao WANG, Xiao-yan CHEN, Xiao-long LEI. Study on Non-Destructive Detection Method of Kiwifruit Sugar Content Based on Hyperspectral Imaging Technology[J]. Spectroscopy and Spectral Analysis, 2021, 41(7): 2188
    Download Citation