Discrimination of Varieties of Yellow Wines by Using Vis/NIR Spectroscopy and PLS-BP Model

The combination of visible/near infrared (Vis/NIR) transmission spectroscopy and partial least squares-artificial neural network (PLS-ANN) model has been employed for the discrimination of varieties of yellow wines. Wines (n=190) were scanned in the visible and NIR region (325-1075 nm) in a monochromator instrument in transmission. The PLS analysis indicated that the accumulative reliabilities of principal components (PCs1-12) were more than 98.34% and the 2-dimentional plot with the scores of PC 1 and PC 2 provided the best clustering of the three varieties of yellow wines. The compressed new variables PCs1-12 were used as the ANN inputs. 145 samples from three varieties were selected randomly to develop the training model, and then using it to validate the 45 unknown samples which were not used in the training data sets. The discrimination ratio of 100% was achieved. It demonstrated the potential use of Vis/NIR spectroscopy combined with PLS-ANN as an available and rapid approach to classify the yellow wines.