Figures (4)  Tables (1)
    • Figure 1. 

      Raw spectra from instruments before pre-processing data treatment: (a) DG-NIR, (b) HSI-NIR, (c) ATR-FTIR, (d) Raman. GCP samples are labelled according to the country of origin (B: Brazil, E: Ethiopia, I: Indonesia, R: Rwanda).

    • Figure 2. 

      (a) Pre-processed DG-NIR spectra, (b) scores, (c) first loading, (d) second loading of (i) MSC with MNCN, (ii) SNV with MNCN, (iii) EMSC with MNCN pre-processed DG-NIR spectra.

    • Figure 3. 

      (a) Pre-processed HSI-NIR spectra measured in reflectance, (b) scores, (c) first loading of (i) Normalisation with MNCN, (ii) SG (1st der, 2nd poly, 15 pts) with MNCN, (iii) Normalisation, SG (1st der, 2nd poly, 15 pts) with MNCN pre-processed HSI-NIR spectra.

    • Figure 4. 

      (a) Pre-processed Raman spectra, (b) scores, (c) first loading of (i) MSC with MNCN, (ii) EMSC with MNCN, (iii) WLS, Normalisation with MNCN pre-processed Raman spectra. Loading plots were produced using R.

    • Optimised pre-processingTVar %RMSECVRMSECRMSEPMCC, Pred.Accuracy, PredF1, Pred.
      DG-NIRMNCN98.540.3380.3290.4730.3830.6650.483
      MSC, MNCN91.490.2450.2430.3090.7740.8820.865
      SNV, MNCN91.490.2450.2430.3090.7740.8820.865
      SNV, Detrend, MNCN91.320.2450.2430.3090.7740.8820.865
      MSC, SG (1st der, 2nd poly, 15 pts), MNCN76.060.2680.2650.3500.6840.8350.788
      Normalisation, SG (2nd der, 2nd poly, 7 pts), MNCN98.050.3580.3520.3510.6520.8120.757
      EMSC, MNCN87.870.2400.2380.2500.8760.9290.916
      HSI-NIRMNCN99.690.3720.3620.4210.6180.8000.681
      Normalisation, MNCN98.280.3330.3220.4020.6550.7670.650
      SG (1st der, 2nd poly, 15 pts), MNCN68.870.3410.3250.3640.6360.8000.728
      MSC, SG (1st der, 2nd poly, 15 pts), MNCN63.410.3380.3240.4030.4730.7330.605
      Normalisation, SG (1st der, 2nd poly, 15 pts), MNCN85.790.3240.3130.3750.6120.8000.732
      SNV, SG( 1st der, 2nd poly, 15 pts), MNCN63.380.3370.3240.4020.4730.7330.605
      FTIRMNCN99.690.3350.3210.3860.2530.4520.200
      Normalisation, MNCN98.170.3340.3200.3910.3720.4520.179
      Normalisation, SG (1st der, 2nd poly, 15 pts), MNCN97.120.4020.3690.4900.1410.5000.330
      EMSC, MNCN71.520.4090.3840.4820.2860.5000.326
      RamanMNCN99.910.3190.3120.3290.7560.8600.819
      SG (2nd der, 2nd poly, 7 pts), MNCN99.660.3500.3430.3690.5210.7350.648
      Normalisation, SG (1st der, 2nd poly, 15 pts), MNCN98.940.3210.3150.3340.5540.7470.622
      WLS (2nd poly), MNCN96.860.3430.3360.3720.6110.7950.760
      FT, Fourier-Transform; DG, Dispersive; HSI, Hyperspectral Imaging; NIR, near-infrared; TVar, Total explained variance; RMSE(C/CV/P), Root Mean Square Errors of Calibration/Cross-Validation/Prediction; MCC, Matthew's Correlation Coefficient, MNCN, Mean centering; MSC; Multiplicative Scatter Correction; EMSC, Extended Multiplicative Scatter Correction; SG (#der, #poly, #pts), Savitzky-Golay #derivative, #polynomial, #window points; WLS, Weighted Least Squares; Pred., Prediction.

      Table 1. 

      Prediction statistics associated with optimal pre-processing methods for spectral data collected using DG-NIR, HSI-NIR, FTIR and Raman.