摘要
采用便携式近红外光谱仪采集231个霍山石斛枫斗和河南石斛枫斗样品的近红外光谱,采用一阶导数(1st D)、标准正态变量变换(SNV)、均值中心化(MNCN)、多元散射校正(MSC)、矢量归一化(N)等方法及其组合的9种方法预处理原始光谱,应用偏最小二乘判别分析(PLS-DA)建立快速无损鉴别霍山石斛枫斗和河南石斛枫斗的数据模型,比较不同光谱预处理方法对PLS-DA建立模型的准确率影响,以预测模型的正确率、敏感性、特异性为指标,评价模型的优劣。结果表明:1st D+SNV+MNCN预处理方法的效果最好,模型的准确率最高,在潜在变量是12的情况下,校正集、交互验证集、预测集的准确率都为100%;模型的敏感性、特异性都为100%。
The near infrared spectra of 231 samples of Dendrobium huoshanense and Dendrobium henanense were measured by portable near infrared spectroscopy. Nine kinds of preprocessing methods,such as first derivative（ 1 st D）,standard normal variate（ SNV）,mean centering（ MNCN）,multiplicative signal correction（ MSC）,normalize（ N） and their combination were used to pretreat original spectrometer. The model of identifying D. huoshanense and D. henanense were established with PLS-DA method.Different spectral preprocessing methods were compared with the accuracy of model. The models were also evaluated with the accuracy,sensitivity and specificity of PLS-DA model. The results showed that the 1 st D ＋ SNV ＋ MNCN preprocessing method had the best effect,the accuracy of the model was the highest. When the latent variable was 12,the prediction accuracy of calibration set,cross validation set and prediction set were 100%,and 100% specificity and sensitivity rate of PLS-DA model were also obtained. The results indicted that PLS-DA analysis with Near-Infrared can provide a new method for rapid and noninvasive identification of D. huoshanense Fengdou and D. henanense Fengdou.
出处
《林产化学与工业》
EI
CAS
CSCD
北大核心
2017年第5期101-106,共6页
Chemistry and Industry of Forest Products
基金
科技部中医药行业专项资助(201407003)
安徽省石斛产业化开发协同创新计划资助(无编号)
关键词
便携式近红外光谱仪
PLS-DA
霍山石斛枫斗
河南石斛枫斗
无损鉴别
portable NIR spectrometer
PLS-DA
Dendrobium huoshanense Fengdou
Dendrobium henanense Fengdou
nondestructive identification
