Abstract. Wine grape variety is one of the main determinants of wine quality. The objective of this study is to explore the feasibility of using hyperspectral imaging (HSI) to identify six red and six white wine grape cultivars during the ripening period. Abnormal spectral data were removed by the Mahalanobis distance, and six different methods were employed to preprocess the spectral data. Next, the effective wavelengths for the classification of grape varieties were selected using principal component analysis (PCA) loadings to improve the HSI processing speed. Finally, three methods were applied to classify grape samples: a support vector machine (SVM), a random forest (RF), and an AdaBoost model. The results indicated that the model established by Savitzky-Golay (S-G) Filter + PCA + SVM achieves the best classification result. The average calibration and validation accuracy for red grapes reached 93.06% and 90.01%, respectively, and for white grapes, they reached 83.77% and 81.09%, respectively, which are slightly lower than those achieved by the full-spectrum model. This study revealed that hyperspectral imaging has great potential for rapid variety discrimination of different wine grapes. Keywords: Hyperspectral imaging, Random forest, Support vector machine, Variety identification, Wine grape.
Using Support Vector Regression and Hyperspectral Imaging for the Prediction of Oenological Parameters on Different Vintages and Varieties of Wine Grape Berries
