Abstract
Wine grape quality is influenced by the variety and growing environment, and the quality of the grapes has a significant impact on the quality of the wine. Tannins are a crucial indicator of wine grape quality, and, therefore, rapid and non-destructive methods for detecting tannin content are necessary. This study collected spectral data of Pinot Noir and Chardonnay using a geophysical spectrometer, with a focus on the 500-1800 nm spectrum. The spectra were preprocessed using Savitzky-Golay (SG), first-order differential (1D), standard normal transform (SNV), and their respective combinations. Characteristic bands were extracted through correlation analysis (PCC). Models such as partial least squares (PLS), support vector machine (SVM), random forest (RF), and one-dimensional neural network (1DCNN) were used to model tannin content. The study found that preprocessing the raw spectra improved the models' predictive capacity. The SVM-RF model was the most effective in predicting grape tannin content, with a test set R(2) of 0.78, an RMSE of 0.31, and an RE of 10.71%. These results provide a theoretical basis for non-destructive testing of wine grape tannin content.