近红外光谱法直接检测甜叶菊叶片甜菊糖苷模型建立

Determination of Steviol in Stevia Rebaudiana Leaves by Near Infrared Spectroscopy

使用近红外光谱技术直接扫描甜叶菊干叶片,建立了甜菊苷(stevioside,ST)和莱鲍迪苷A(rebaudioside A,RA)的检测模型。对甜菊苷含量在0.27%~1.40%,莱鲍迪苷A含量在0.61%~3.98%范围内的不同品种的甜叶菊干叶片进行了近红外光谱扫描,共扫描了105份。采用偏最小二乘法建立甜菊糖苷的检测模型,比较了减去一条直线、多元散射校正、一阶导数和二阶导数等不同的光谱预处理方法对模型的影响。结果显示减去一条直线的数据预处理方法为ST的最优建模方法。ST校正集相关系数为0.986,校正均方根误差为0.341,预测均方根误差为1.00,相对分析误差为2.8;RA采用无光谱预处理建模,RA的建模结果相关系数为0.967,校正均方根误差为1.50,预测均方根误差为1.98,相对分析误差为4.17。说明近红外光谱技术检测甜叶菊干叶片中ST和RA的含量具有一定的可行性。同时与甜叶菊粉末ST模型结果相关系数为0.986,校正均方根误差为0.32,预测均方根误差为0.601,相对分析误差为2.86和RA模型结果相关系数为0.968,校正均方根误差为1.50,预测均方根误差为1.48,相对分析误差为4.2相比差异不明显。但减少了叶片粉末检测过程中的烘干、研磨的步骤,节省了时间,降低了工作量。

The objective of the present study is to develop a method for rapid determination of the content of stevioside (ST) and rebaudioside A (RA) in Stevia Rebaudiana leaves .One hundred and five samples of stevia from different areas containing ST of 0.27% ~1.40% and RA of 0.61% ~3.98% were used .The 105 groups’ NIRS diagram was processed by different methods in-cluding subtracting a straight line (SLS) ,multiplicative scatter correction (MSC) ,first derivative (FD) ,second derivative (SD) and so on ,and then all data were analyzed by partial least square (PLS) .The study showed that SLS can be used to extracted spectra information thoroughly to analyze the contents of ST ,the correlation coefficients of calibration (Rc ) ,the root-mean-square errors of calibration (RMSEC) and prediction (RMSEP) ,and the residual predictive deviation (RPD) were 0.986 , 0.341 ,1.00 and 2.8 ,respectively .The correlation coefficients of RA was 0.967 ,RMSEC was 1.50 ,RMSEP was 1.98 and RPD was 4.17 .The results indicated that near infrared spectroscopy (NIRS) technique offers effective quantitative capability for ST and RA in Stevia Rebaudiana leaves .Then the model of stevia dried leaves was used to compare with the stevia powder near infrared model whose correlation coefficients of ST was 0.986 ,RMSEC was 0.32 ,RMSEP was 0.601 and RPD was 2.86 and the correlation coefficients of RA was 0.968 ,RMSEC was 1.50 ,RMSEP was 1.48 and RPD was 4.2 .The result showed that there was no significant difference between the model of dried leaves and that of the powders .However ,the dried leaves NIR model reduces the unnecessary the steps of drying and grinding in the actual detection process ,saving the time and reducing the workload .