基于NIR和化学指标的国产烤烟烟叶产地、部位模式识别

Pattern Recognition of Growing Area and Stalk Position of Domestic Flue-cured Tobacco Based on NIR and Chemical Components

分别以NIR光谱和总糖、还原糖、烟碱、总氮、钾、氯检测数据为基础,采用基于马氏距离的判别法对2003、2004年1129个国产烤烟烟叶样品的产地和部位进行了模式识别。结果表明:采用NIR光谱与这6个主要化学成分指标对烟叶产地的识别准确率分别为88%~94.5%和62%~78%,对不同产地烟叶部位的识别准确率分别为71%~75%与63%~67%,对同一产地烟叶部位的识别准确率分别为82%~87%与80%~93%,对上、下部烟叶模式识别的识别准确率分别为92%~98%与89%~98%。NIR光谱可用于烟叶产地、部位的识别,这6个主要化学成分指标仅适合于烟叶部位的识别。

The pattern recognition of growing area and stalk position of 1129 domestic flue-cured tobacco samples in 2003 and 2004 were carried out by Mahalanobis distance-based discrimination dependent on their near infrared spectra （NIRs） and main chemical components, including total sugar, reducing sugar, nicotine, total nitrogen, potassium and chlorine. The results indicated that the accuracies of growing area recognition by NIRs and chemical components were 88% to 94.5% and 62% to 78% , respectively; and the accuracies of stalk position recognition for tobacco from different areas were 71% to^75% and 63% to 67% , respectively, and for tobacco from the same area were 82% to 87% and 80% to 93% , respectively; the accuracies of upper and lower leaves recognition were 92% to 98% and 89% to 98% , respectively. It was concluded that NIR was suitable to the recognition of growing area and stalk position, while the six main chemical components were only fit for the recognition of stalk position.