Table 3.
Comparison of different spectral pre-processing methods used to develop the calibration models. SNV: standard normal variant, MSC: multiplicative scatter correction, SG: second derivative spectra (Sativzky-Golay method).
| Model number | Data pre-treatment | Latent variables | R2 training | RMSE (%) training | R2 test | RMSE (%) test |
|---|---|---|---|---|---|---|
| 1 | None | 4 | 0.9974 | 0.6538 | 0.9115 | 2.4569 |
| 2 | MSC | 4 | 0.9942 | 0.9802 | 0.8708 | 2.9692 |
| 3 | SNV | 4 | 0.9994 | 0.2994 | 0.9643 | 1.5610 |
| 4 | SG | 4 | 0.9967 | 0.7397 | 0.9833 | 1.0662 |
| 5 | MSC + SG | 4 | 0.9956 | 0.8523 | 0.9360 | 2.0891 |
| 6 | SNV + SG | 4 | 0.9977 | 0.6145 | 0.9516 | 1.8174 |