摘要
采用傅里叶变换红外光谱(Fourier Transform Infrared Spectroscopy,FT-IR)仪结合衰减全反射(Attenuated Total Reflection,ATR)附件采集石斛茎部(近根部)横断面的中红外光谱,原始光谱经标准正态变换(Standard Normal Variate,SNV)和均值中心化(Mean Center,MNCN)预处理后,采用偏最小二乘法判别分析法(Partial Least Squares Discriminant Analysis,PLSDA)建立两种石斛的鉴别模型。结果显示,全谱PLSDA方法所建模型校正集、校正集交叉验证和预测识别率分别为96.25%、92.69%和91.82%。采用无信息变量消除法(Uninformative Variable Elimination,UVE)优选敏感波长后,建立PLSDA模型的准确性更高,校正集、校正集交叉验证和预测集识别正确率分别达到了99.28%、95.72%和95.02%。
We collected spectra of dendrobium at the stem near root by Fourier transform attenuated total reflection （ATR） . The raw spectra were preprocessed by standard normal variate （SNV） and mean center（MNCN）,using partial least squares discriminant analysis （PLSDA） to build model. Results demonstrated that the prediction precision of modll buitt with full wavelength variables was good , the forecast recognition rate in calibration set , cross validation for calibration set and prediction set were 96. 25%,92. 69% and 91. 82% respectively. After wavelength variables optimized by UVE,forecast recognition rate in calibration set,cross validation for calibration set and prediction set were improved to 99. 28%, 95. 72% and 95. 02% respectively.
出处
《皖西学院学报》
2017年第5期1-5,共5页
Journal of West Anhui University
基金
安徽省教育厅重点项目"基于近红外光谱技术的石斛类药材快速无损识别研究"(KJ2014A279)
安徽省石斛产业化开发协同创新中心计划
关键词
霍山石斛
河南小石斛
傅里叶变换红外光谱
偏最小二乘法判别分析法
无信息变量消除
Dendrobium Huoshanense
Dendrobium Henan
Fourier transform infrared spectroscopy
partial leasdiscriminant analysis
uninformative variable elimination
