Nie M.,Hunan University |
Zhou J.,Hunan University |
Yang R.,Yunnan Qujing Tobacco Company |
Xia K.,Yunnan Qujing Tobacco Company |
And 2 more authors.
Acta Tabacaria Sinica | Year: 2014
Support vector machine (SVM) which was used to predict smoking quality based on chemical index of flue-cured tobacco was established by using libsvm software package in Matlab environment. Volatile acid, extractable petroleum ether, potassium, reducing sugar and nicotinamide were taken as input variables by using MIV algorithm. Optimal C value was 0.10882, g value was 9.1896 and correlation coefficient R was 0.9903 for training samples and 0.9966 for testing samples. Predicted error was distributed in the [-4, 4]. Results showed that the prediction model is reliable and capable of providing reference for smoking quality evaluation. ©, 2014, State Tobacco Monopoly Bureau and China Tobacco Society. All right reserved.
Tang G.,China Agricultural University |
Tian K.,China Agricultural University |
Li Z.,Yunnan Qujing Tobacco Company |
Zheng B.,Yunnan Qujing Tobacco Company |
Min S.,China Agricultural University
Tobacco Science and Technology | Year: 2013
In order to rapidly classifying tobacco grades, the classification model of tobacco grades was established by partial least squares discrimination analysis (PLS-DA) with the near-infrared spectra of 150 tobacco samples collected from Qujing in Yunnan. The prediction were carried out on the 60 samples in the prediction set with the established model. The results indicated that: 1) The prediction accuracies of training and prediction sets were 100.0% (150/150) and 96.7% (58/60), respectively; 2) PLS-DA was efficient in tobacco grade classifying. This model provides a new rapid discrimination analysis method for the classification of tobacco grades.
Wu L.-J.,China Agricultural University |
Tian K.-D.,China Agricultural University |
Li Q.-Q.,China Agricultural University |
Li Z.-H.,Yunnan Qujing Tobacco Company |
And 2 more authors.
Guang Pu Xue Yu Guang Pu Fen Xi/Spectroscopy and Spectral Analysis | Year: 2013
In the present paper, the authors used NIR to determine routine chemical components, namely total sugar, reducing sugar, nicotine, total nitrogen, starch and volatile alkali. Orthogonal signal correction (OSC) was employed as spectral pretreatment and the principal component regression (PCR) models for 6 chemical components were established with Monte Carlo cross-validation modeling strategy. RPD value for each model was calculated to evaluate the methods. The orientation of PCR projection is the largest variance direction and has no relationship with the concentration. OSC can not only get rid of uninformative concentration but also solve the problem of noise, baseline drift and stray light. Compared with conventional PCR, OSC-PCR sustains the accuracy of the predicting model and improves the stability of the model significantly. It proves that NIR coupled with OSC-PCR method can be applied to the determination of routine chemical components, which is of great significance in evaluation of tobacco quality and analysis of tobacco aroma components.