基于近红外光谱的烟叶SIMCA模式识别

研究了基于烟叶的近红外光谱数据通过软独立模式分类(SIMCA)识别不同烟叶的方法。首先对每种具有确定产地、等级、品种的目标烟叶进行多次分布式取样，扫描目标烟叶多个样品的近红外光谱；再对目标烟叶近红外光谱进行主成分分析(PCA)运算生成每种目标烟叶的数据模型；然后扫描未知烟叶的近红外光谱，用目标烟叶数据模型对未知烟叶近红外光谱进行主成分分解计算，计算未知烟叶与目标烟叶的距离，通过距离衡量未知烟叶与目标烟叶的相似程度。建立了包含115种不同产地、等级、品种的目标烟叶的数据模型，对115个外部检验样品进行了模式识别，正确识别率高于90%。结果表明该烟叶模式识别方法基础数据易得，同时考虑了烟叶的平均水平和分布水平，识别准确率高，具有良好的发展前景。 

SIMCA pattern recognition of tobacco leaves based on near infrared spectra

Soft independent modeling of class analogy (SIMCA) method based on near infrared spectra was applied for the recognition of various tobacco leaves. NIR spectra of samples of each destination tobacco leaf were collected. PCA model of specific destination tobacco was built by principal component analysis on the NIR spectra of the destination tobacco. Distances of spectrum of unknown tobacco leaf from PCA models of all destination tobacco leaves were calculated. The distance represented similarity of unknown tobacco leaf and the destination tobacco leaves. Models of 115 destination tobacco leaves of different production area, grade, and variety were built and 115 independent validation tobacco samples were classified by these models. The ratio of correct recognition was over 90%.