Analysis of Tobacco by Near-infrared Spectroscopy and Support Vector Machine

In this study, in order to establish analysis models of near-infrared spectroscopy(NIR) of tobacco, 120 samples of tobacco from different cultivation area were surveyed by near-infrared( NIR) spectroscopy. As the new pattern recognition, support vector machine(SVM) can avoid over-fitting problem and owns the superior generalization ability and prediction accuracy, were applied in this study. The quantitative and qualitative analysis models of tobacco samples were studied separately in this experiment using radial basis function (RBF) SVM. For reducing dimension and moving noise, the spectrum variables were highly effectively compressed using the wavelet transformation(WT) technology and the haar wavelet was selected to decompose the spectroscopy signals. Simultaneously, the parameters of the models were also discussed in detail. The best experimental results were obtained using the RBF SVM regression with kernel parameter sigma = 1.0, 1.2, 1.4, 0.6, separately corresponds to total-sugar, reducing sugar, nicotine, total-nitrogen, and RBF SVM classifier with kernel parameter sigma = 1.6. Meanwhile, the values of appraisal index, namely coefficient of determination (R-2), root mean squared error of prediction(RMSEP) and mean relative error(RME), indicate its excellent generalization for quantitative and qualitative analysis results and high prediction accuracy. The overall results show that NIR spectroscopy combined with SVM can be efficiently utilized for rapid and accurate analysis of chemical compositions in tobacco and discrimination of tobacco of different origins. On the other hand, the research can show that SVM is effective modeling tools to NIR spectroscopy and can provide technical support for quantitative and quantitative analysis of other NIR applications.