基于近红外漫反射光谱法,结合偏最小二乘法建立了甜叶菊中甜菊糖苷总量(TSG)、瑞鲍迪苷A (RA)、甜菊苷(STV)、绿原酸总量及水分的定量分析模型。选取不同地区的500个甜叶菊样品,以高效液相色谱法(TSG、RA、STV、绿原酸总量)和烘干法(水分)所得数据为参比,结合样品的近红外光谱图,按照以下条件进行建模:1 TSG模型,校正集385,验证集97,光谱预处理采用多元散射校正(MSC)+一阶导数(1st)+Norris derivative滤波平滑(ND),光谱范围4 090.76~7 085.37 cm^-1,主因子数8;2 RA模型,校正集381,验证集94,光谱预处理采用MSC+二阶导数(2nd)+ND,光谱范围4 060.38~6 221.23 cm^-1、6 769.51~7 401.24 cm^-1,主因子数7;3 STV模型,校正集386,验证集96,光谱预处理采用MSC+1st+ND,光谱范围4 017.86~4 224.39 cm^-1、4 370.17~5 172.15 cm^-1、5 414.95~9 106.22 cm^-1,主因子数5;4绿原酸总量模型,校正集376,验证集95,光谱预处理采用标准正态变量变换(SNV)+1st+Savitzky-Golay卷积平滑(SG),光谱范围4 000.21~5 300.00 cm^-1、5 624.00~6 246.90 cm^-1、8 746.50~9 373.80 cm^-1,主因子数12;5水分模型,校正集389,验证集96,光谱预处理采用MSC+1st+ND,光谱范围4 072.53~7 553.09 cm^-1,主因子数8。结果显示:TSG、RA、STV、绿原酸总量和水分模型的校正相关系数、预测相关系数、交叉验证相关系数均大于0.800 0,校正均方根误差、预测均方根误差、交叉验证均方根误差均小于0.500;对各模型进行外部验证,TSG、RA、STV、绿原酸总量和水分的预测值与实测值的拟合相关系数均大于0.880 0;利用模型对甜叶菊样品中TSG、RA、STV、绿原酸总量和水分进行分析,日内精密度(n=6)为0.54%~2.7%,日间精密度(n=6)为1.1%~4.7%。模型用于某试验基地不同生长批次甜叶菊中TSG、RA、绿原酸总量的测定,TSG质量分数为10.40%~13.32%,RA质量分数为4.99%~7.61%,绿原酸质量分数为2.73%~4.07%,测定值的相对标准偏差(n=12)均小于7.0%。

Based on near infrared diffuse reflection spectrometry, the models of total steviol glucosides (TSG), rebaudioside A (RA), stevioside (STV), total chlorogenic acids and moisture in stevia combined with partial least square (PLS) method were established. Stevia samples of 500 from different regions were selected, and the data obtained by high performance liquid chromatography (TSG, RA, STV, total chlorogenic acids) and oven drying method (moisture) were used as reference. The models were established combined with near infrared spectra of samples according to the following conditions: 1 TSG model, calibration set of 385, verification set of 97, spectral pretreatments of MSC+1st+ND, spectral range of 4 090.76-7 085.37 cm^-1, and principal factor number of 8; 2 RA model, calibration set of 381, verification set of 94, spectral pretreatments of MSC+2nd+ND, spectral ranges of 4 060.38-6 221.23 cm^-1, 6 769.51-7 401.24 cm^-1, and principal factor number of 7; 3 STV model, calibration set of 386, verification set of 96, spectral pretreatments of MSC+1st+ND, spectral ranges of 4 017.86-4 224.39 cm^-1, 4 370.17-5 172.15 cm^-1, 5 414.95-9 106.22 cm^-1, and principal factor number of 5; 4 total chlorogenic acids model, calibration set of 376, verification set of 95, spectral pretreatments of SNV+1st+SG, spectral ranges of 4 000.21-5 300.00 cm^-1, 5 624.00-6 246.90 cm^-1, 8 746.50-9 373.80 cm^-1, and principal factor number of 12; 5 moisture model, calibration set of 389, verification set of 96, spectral pretreatments of MSC+1st+ND, spectral ranges of 4 072.53-7 553.09 cm^-1, and principal factor number of 8. It was shown that the correlation coefficients of correction, prediction and cross validation for TSG, RA, STV, total chlorogenic acids and moisture models were all greater than 0.800 0. The root mean square errors of correction, prediction and cross validation were less than 0.500. External validation of the models showed that the correlation coefficients between predicted values and determined values for TSG, RA, STV, total chlorogenic acids and moisture were greater than 0.880 0. TSG, RA, STV, total chlorogenic acids and moisture in stevia samples were analyzed by their models, and the intra-day precision (n=6) and inter-day precision (n=6) were in the ranges of 0.54%-2.7% and 1.1%-4.7%, respectively. The models have been applied to the determination of TSG, RA and total chlorogenic acids in stevias from different growth batches at a test base, and the mass fractions of TSG, RA and total chlorogenic acids were in the ranges of 10.40%-13.32%, 4.99%-7.61% and 2.73%-4.07%, respectively, with RSDs (n=12) of the determined values were less than 7.0%.