文献类型：期刊文献

中文题名：不同产地国产烤烟近红外光谱的特征分析及其模式识别

英文题名：NIR Characteristic Analysis and Pattern Recognition of Domestic Flue-cured Tobaccos

作者：王国东[1];束茹欣[2];张建平[2];倪力军[1]

机构：[1]华东理工大学化学与制药学院;[2](上海烟草(集团)技术中心,上海市许昌路1062号200082

年份：2006

卷号：39

期号：5

起止页码：36

中文期刊名：烟草科技

外文期刊名：Tobacco Science & Technology

收录：CSTPCD;;Scopus;北大核心:【北大核心2004】；

语种：中文

中文关键词：烤烟;近红外光谱;主成分分析;光谱预处理;模式识别

外文关键词：Flue-cured tobacco ; Near infrared spectra ; Principal component analysis ; Spectral pretreatment ; Pattern recognition

摘要：对2003年125个不同产地的国产烤烟烟叶的原始近红外光谱、SNV光谱与一阶导数光谱进行了主成分分析,采用主成分空间下的马氏距离判别样本的产地归属,提出根据仪器噪声和误差水平确定最大主成分个数,研究了主成分个数、样本空间、光谱区间对烟叶产地识别准确率的影响,分析了烟叶产地的近红外特征区和产地特征信息在不同主成分上的体现。结果表明,采用光滑处理的全光谱区间的一阶导数光谱,在建模样本空间进行主成分分析时所建模型对烟叶样品的识别准确率最高。
By studying the original, SNV and first derivative near infrared spectra （NIRS） of 125 domestic flue-cured tobacco samples from different areas in 2003, the principal components in these samples were analyzed. The grown areas of tobacco samples were discriminated by Mahalanobis distance under principal components＇ space. It was proposed to determine the maximum number of principal component on the basis of instrument noise and error level. The effects of principal component number, specimen space and spectral region on the accuracy in discrimination of tobacco grown area were studied. The embodiments of NIR characteristic regions and characteristic informations of grown area on different principal components were analyzed. The results indicated that the highest recognition accuracy was offered by the models which were established while analyzing principal components in modeling specimen spaces and the first derivative near infrared spectra in full wavelength range were used.