为了探讨不同产地、不同等级以及不同品种来料烟外观指标、物理特性及近红外光谱对其感官质量的影响，找到一种可以客观反映烟叶感官质量的方法。选取了四川3个产地、3个部位、10个等级和2个品种的单料烟，以组为单位进行采集，每组10片，共采集烟叶7897片。使用变量标准化(SNV)方法对原始光谱数据进行校正，用Standard Scaler方法对烟叶外观指标、物理特性、校正后的光谱数据进行标准化。采用主成分分析(PCA)法对数据进行降维。分别运用二次判别分析算法(QDA)、K近邻算法(KNN)、支持向量机(SVC)构建训练模型，用Grid Search CV算法进行超参优化，以平衡准确率作为模型评价指标。结果表明，3个训练模型中，SVC的泛化能力最优。其中，香气风格彰显程度、香气质、香气量、杂气、浓度、劲头、刺激性、余味及甜感的预测平衡准确率分别为0.747、0.751、0.715、0.720、0.712、0.774、0.685、0.725、0.700。外观指标、物理特性及近红外光谱共同影响着烟叶的感官质量。

To investigate the influence of the appearance index, physical properties and near-infrared spectroscopy of cigarettes from different origins, different grades and different varieties on the sensory qualityand to find a way that could objectively reflect the sensory quality of tobacco leaf, in this study, single tobaccos including 3 parts, 10 grades and 2 varieties were selected from 3 origins of Sichuan Province, and 10 pieces of tobacco leaves in each group were adopted for the collection, i.e. a total of 7897 tobacco leaves were collected. The original spectral data was corrected by standard normalized variate (SNV), and the appearance index, physical properties and standarded spectral data of tobacco leaf were standardized with the Standard Scaler method subsequently. Principal Component Analysis(PCA) was used to reduce the dimensionality of the data. The quadratic Discriminant Analysis Algorithm(QDA), K-Nearest Neighbors (KNN), and Support Vector Machine (SVC) were performed to construct the training model, and the Grid Search CV algorithm was executed for hyperparameter optimization, and the balance accuracy was used as the model evaluation. The results showed that mong the three training models, SVC had the best generalization ability. The prediction balance accuracy of aroma style highlighting, aromatric, aroma volume,miscellaneous gas, concentration, strength, irritation, aftertaste and sweetness were 0.747, 0.751, 0.715, 0.720, 0.712, 0.774, 0.685, 0.725, 0.700, respectively. The appearance index, physical properties and near-infrared spectra jointly affect the sensory quality of tobacco leaf.

S-3

四川中烟工业有限责任公司项目(KJSB202104150020)。