Rapid Discrimination of Base Liquor for Baijiu Based on FT-NIR Spectroscopy and KPCA-MD-SVM
Article
Figures
Metrics
Preview PDF
Reference
Related
Cited by
Materials
Abstract:
To ensure the accuracy of base liquor segmentation in the process of liquor picking, the spectra of base liquor samples during the whole liquor picking process were collected using Fourier transform near-infrared spectroscopy, and support vector machines were used to establish a segmentation model for the spectrum of the base liquor with the optimal pretreatment. The accuracy of the model training set was 93.02%, and the discrimination rate for the test set was 90.08%. To reduce the modeling time and increase the reliability of the model, kernel principal component analysis was used to reduce the dimensionality of the base liquor spectral data. A segmentation model for the base liquor was established. The accuracy of the model for the training set was 94.81%, and the discrimination rate for the test set was 90.75%, which were 1.79% and 0.67% higher than those of the model without kernel principal component analysis, respectively. To further improve the discrimination ability of the model, the Mahalanobis distance was used to eliminate abnormal data samples after dimensionality reduction. The accuracy of the new model for the training set in terms of the number of base liquor segments was 98.72%, and the accuracy for the test set was 98.75%. The accuracy for the training set and discrimination rate for the test set were improved by 3.91% and 8%, respectively. Thus, the kernel principal component analysis + Mahalanobis distance + support vector machine-based liquor segmentation model can quickly distinguish base liquor, suggesting that near infrared spectroscopy can be applied in automatic liquor picking.
