CAJ | 학술논문

应用拉曼光谱结合化学计量学方法对蜂蜜果糖和葡萄糖含量进行了定量分析。用自适应迭代重加权惩罚最小二乘（adaptive iteratively reweighted penalized least squares，airPLS）算法进行基线校正，用竞争性自适应重加权采样（competitive adaptive reweighted sampling，CARS）算法筛选变量，分别用线性的偏最小二乘（partial least squares，PLS）回归算法和非线性的支持向量机（support vector machines，SVM）回归算法建立定量校正模型，并进行预测。2种模型都有较好的预测结果。对果糖，SVM模型预测值与高效液相色谱法（high performance liquid chromatography，HPLC）测定值的相关系数（R）和预测均方根误差（root mean square error of prediction，RMSEP）分别为0.902和1.401，略优于PLS模型（R为0.892，RMSEP为1.604）；对葡萄糖，PLS模型的R和RMSEP分别为0.968和0.669，优于SVM模型（R为0.933，RMSEP为1.410）。结果表明拉曼光谱结合化学计量学方法可快速无损测定蜂蜜果糖和葡萄糖含量。
응용랍만광보결합화학계량학방법대봉밀과당화포도당함량진행료정량분석。용자괄응질대중가권징벌최소이승（adaptive iteratively reweighted penalized least squares，airPLS）산법진행기선교정，용경쟁성자괄응중가권채양（competitive adaptive reweighted sampling，CARS）산법사선변량，분별용선성적편최소이승（partial least squares，PLS）회귀산법화비선성적지지향량궤（support vector machines，SVM）회귀산법건립정량교정모형，병진행예측。2충모형도유교호적예측결과。대과당，SVM모형예측치여고효액상색보법（high performance liquid chromatography，HPLC）측정치적상관계수（R）화예측균방근오차（root mean square error of prediction，RMSEP）분별위0.902화1.401，략우우PLS모형（R위0.892，RMSEP위1.604）；대포도당，PLS모형적R화RMSEP분별위0.968화0.669，우우SVM모형（R위0.933，RMSEP위1.410）。결과표명랍만광보결합화학계량학방법가쾌속무손측정봉밀과당화포도당함량。
Raman spectroscopy combined with chemometric methods was used to rapidly measure the content of fructose and glucose in honey. Seventy-five authentic honey samples from sixteen floral origins were obtained directly from bee-keepers in ten provinces of China from 2008 to 2010. The samples were stored at 6-8℃in the laboratory before their analysis. Honey were liquefied in a water bath at 55℃ and manually stirred to ensure homogeneity before spectral measurements. Spectra of honey samples were recorded using an i-Raman spectrometer (BWS 415-785H, B&W TEK Inc., USA), which was equipped with a fiber-optic Raman probe, a thermoelectric cooled CCD detector with 2048 pixels and a 785 nm laser with a maximum output power of 495 mW in the signal range of 175-2 600 cm-1. The instrumental spectral resolution was 3 cm-1. Integration time was 15 s. Seventy-four samples were divided into 55 calibration sets and 19 validation sets by Kennard-Stone algorithm. AirPLS (adaptive iteratively reweighted penalized least squares) was used to correct the baseline of spectroscopy. CARS (competitive adaptive reweighted sampling) was used to screen variables. Thirty-one and forty-six variables were obtained from 1150 variables by CARS for glucose and fructose, respectively. Quantitative calibration models were developed with linear partial least squares (PLS) regression and non-linear support vector machine (SVM) regression, respectively. These models were used to predict the validation set samples. The prediction accuracies obtained from both glucose and fructose were satisfied by PLS model and SVM model. Correlation coefficient (R)of predicted values versus HPLC measured values and root mean square error of prediction (RMSEP) were 0.902 and 1.401 obtained from SVM model for fructose, respectively, which were higher than the values obtained by PLS model (R=0.892, RMSEP=1.604). PLS model’s R and RMSEP were 0.968 and 0.669 for glucose, respectively, which were higher than SVM model’s values (R=0.933, RMSEP=1.410). Raman spectroscopy combined with chemometric methods is a rapid and non-destructive method, which can be applied to measure the content of fructose and glucose in honey.