国产烤烟烟叶的NIRS模式识别

以近红外光谱的主成分描述烟叶特征,采用马氏距离判别准则对不同产地的国产烤烟烟叶进行了产地、部位、等级的模式识别。结果表明:①预测准确率随样本的复杂程度和数量而变,波动范围为74%~97%。错分率高于80%的样本主要集中在自然生态环境多样性和复杂性程度较高的云南、贵州、四川及重庆地区的交界地域,说明按照行政区划分烟叶产区不尽合理;②部位、等级识别的准确率低于产地的识别,这与样本数不够大、烟叶分级受人为因素影响较大有关。

NIRS-based Pattern Recognition of Domestic Flue-cured Tobacco

The pattern recognition of domestic flue-cured tobaccos grown in different areas in terms of producing area,stalk position and grade was carried out by Mahalanobis distance criterion on the basis of principal components of the leaves characterized by near infrared spectra(NIRS).The results indicated that: 1) the prediction accuracy ranged from 74% to 97%,depended upon the complexity and number of tobacco samples.The samples whose rate of misidentification happened to be over 80% were mainly from the border between Yunnan,Sichuan,Guizhou and Chongqing,where the biological environments were more complex and versatile.Therefore,it challenged the rationality of classifying tobacco producing areas according to administrative demarcations;2) the prediction accuracy of stalk position and grade of tobacco was lower than that of producing area,which related to that the sample number was not enough and leaf grading was substantially affected by man-made factors.