Analysis the Proportion of Cigarette Tobacco Formula by Near Infrared Spectroscopy with Pattern Recognition Method

In order to lay a foundation for cigarette formula substitution and product quality stability evaluation, the recognition model of cigarette tobacco formula proportion was established based on near infrared spectroscopy with pattern recognition method. 5 different proportions of A module tobacco were added into a brand cigarette finished tobacco, and its near infrared spectral information were collected. Near infrared spectra of the samples were pretreated by derivation method (first-order derivative and second-order derivative) and smoothing method (Savitzky-Golay smoothing and Norris smoothing). Combined with principal components analysis-mahalanophil distance (PCA-MD), partial least square method-discriminant analysis (PLS-DA) and orthogonal partial least square method-discriminant analysis (OPLS-DA), the recognition models of the above 5 kinds of finished tobacco were established. It was shown that the best spectral pretreatment method was first-order derivative + Savitzky-Golay smoothing, and the best pattern recognition method was OPLS-DA. When the number of principal components was 4, the cumulative interpretation ability of spectral variables for the optimal recognition model was 0.995, with the cumulative interpretation ability of classified variables of 0.953, the eigenvalue of 0.196, the cumulative crossover validity of 0.192, and the overall recognition rate of external validation for the model was 99%. The results of substitution verification showed that the model was stable and reliable without overfitting phenomenon. The sensory evaluation of 5 finished tobacco was carried out, and the recognition effect of different proportions of cigarette tobacco formula was better by near infrared spectroscopy combined with pattern recognition method.