Detection of SSC in 'Shine Muscat' grapes using Vis/NIR and NIR spectra with characteristic wavelengths

To offer a method for predicting the soluble solids content (SSC) of 'Shine Muscat' with green color in whole ripening period, the characteristic wavelengths which were sensitive to SSC were studied using visible/near-infrared (Vis/NIR) and near-infrared (NIR) spectra techniques. Firstly, the spectral preprocessing methods, including the Savitzky-Golay smoothing filter (SG), standard normal variable transformation (SNV), multiplicative scatter correction (MSC), normalization (Nor), and their combinations were used to remove noise. Subsequently, feature wavelength extraction was performed on the full spectra using uninformative variable elimination (UVE), competitive adaptive reweighted sampling (CARS), genetic algorithm (GA), and successive projections algorithm (SPA), respectively. To reduce the number of selected characteristic wavelengths further, SPA was also employed after UVE, CARS and GA. The results indicate that the partial least squares regression (PLSR) model built using 17 wavelengths selected by the UVE-SPA method in the Vis/NIR spectra achieved optimal performance, with coefficient of determination correlation for the independent test set (R_T²) and residual predictive deviation for the independent test set (RPDT) of 0.734 and 1.922 respectively. In the NIR spectra, the model built using 16 wavelengths selected by the CARS-SPA method demonstrated the best performance with R_T² at 0.923 and RPDT at 3.615. These findings demonstrate that appropriate spectral preprocessing, outlier removal, and feature wavelength selection can reduce model complexity while maintaining prediction accuracy for SSC of 'Shine Muscat' grape. This study provides a theoretical basis for developing handheld non-destructive detectors for SSC of grape.