Prediction of Soluble Solids Content for Wine Grapes During Maturing Based on Visible and Near-Infrared Spectroscopy

The maturity of wine grape is an important quality index to determine the harvest time of grape. Aiming at the problem that the maturity of wine grapeis difficult to be detected in the field, the internal relationship between SSC and spectral data of wine grape was studied by Vis/NIR spectroscopy and chemometrics. The Vis/NIR spectral data of five varieties of grape and their leaves in different mature periods were obtained by USB2000+ spectrometer. The spectral data were extracted by OM-NIC 8. 0 software, and the chemical values and spectral absorption values were modeled by TQ analyst 8. 0 software. The wave band 450 similar to 1 000 nm which had high signal-to-noise ratio was selected, and PCA was adopted to eliminate the abnormal spectral data. The first derivative (FD), Savitzky-Golay smoothing (S-G), multiple scattering correction (MSC) and standard normal variate (SNV) were combined into four methods to preprocess the spectral data. Based on the spectral data of five varieties of grape berry and the spectral data of five varieties of grape leaf, the prediction models of SSC were established by PLS. The model effects with different pretreatment methods were compared, and the optimal pretreatment method was selected for modeling. Finally, the prediction models of SSC were verified by external samples. The results show that the performance of most prediction models is the best when S-G smoothing +FD+FDA+MSC preprocessing method is applied. The correlation coefficient of calibration sets and validation sets of grape berries were above 0. 93 and 0. 86, respectively, and the maximum root means square error is 0. 30 and 0. 48, respectively. The correlation coefficient of calibration sets and validation sets of grape leaves were above 0. 73 and 0. 65, respectively, the maximum root mean square error is 0. 95 and 0. 75, respectively. The highest average relative error between the predicted value and the real value of grape berry samples was 0. 43%. The SSC prediction model built by the spectra of grape berry has a good predictive ability, which is superior to the SSC prediction model built by the spectra of the grape leaf. The prediction model of SSC can provide a theoretical reference for the study of grape maturity evaluation. Therefore, Vis/NIR spectroscopy is suitable for rapid and non-destructive detection of solid soluble content in the wine grape field.