Original Title: Hyperspectral Imaging Spectroscopy for Non-Destructive Determination of Grape Berry Total Soluble Solids and Titratable Acidity
Lyu et al. Remote Sens. 2024, 16(10), 1655; https://doi.org/10.3390/rs16101655
The quality of wine grapes plays a crucial role in determining their market value. Hyperspectral imaging offers a promising, non-destructive approach to predicting key enological parameters. This study explores its feasibility with machine learning models for measuring total soluble solids and titratable acidity in grape berries. A new spectral preprocessing method based on the normalized difference spectral index was developed and compared to the conventional multiplicative scatter correction and Savitzky–Golay smoothing (MSC+SG).
Results showed that the normalized difference spectral index method outperformed MSC+SG in multiple classification models, with the best model correctly classifying 93.8% of total soluble solids and 84.4% of titratable acidity. Additionally, total soluble solids prediction achieved moderate accuracy using support vector regression and MSC+SG (RMSE = 0.523 °Brix, R² = 0.622), while titratable acidity prediction performed similarly with support vector regression and normalized difference spectral index preprocessing (RMSE = 0.19%, R² = 0.525). These findings highlight the advantages of normalized difference spectral index over traditional preprocessing methods for improving predictive accuracy.
Overall, this study demonstrates that hyperspectral imaging, combined with advanced preprocessing techniques like normalized difference spectral index, has strong potential for assessing and grading wine grapes. This approach could enhance quality control in viticulture, helping producers select the best grapes for high-quality wine production.
Would you like to know more? Check it out Remote Sensing, MDPI
Prompted by SmartControl, Powered by ChatGPT and Dall-E