基于近红外特征变量筛选对火麻油掺杂的快速检测
Fast detecting the adulteration of hemp seed oil based on characteristic variables optimization of NIR spectroscopy

利用近红外光谱技术对掺杂了大豆油、花生油、葵花籽油和玉米油的火麻油进行鉴定，结合偏最小二乘法（PLS）和最小二乘支持向量机（LS-SVM）模型建立定量分析，并利用连续投影算法（SPA）和竞争自适应重加权采样算法（CARS）提取特征变量. 结果表明: LS-SVM回归模型的准确度优于PLS模型，其预测相关系数R2p分别达到0.9504、0.9058、0.8574和0.7673；SPA和CARS是两种有效的特征变量选择算法，能够提高模型的准确性，并且CARS效果优于SPA；其中，LS-SVM-CARS模型的R2p分别达到0.9821、0.9075、0.9587和0.9249. 因此，在油脂掺杂快速检测中，LS-SVM-CARS是一个准确度高、变量数少、传递性较强的定量分析模型.
In this paper，near-infrared spectroscopy (NIRs) was used to quantitative analyze the hemp seed oil adulterated with soybean oil，peanut oil，sunflower oil and corn oil. The partial least squares (PLS) and least squares support vector machine (LS-SVM) were applied to analyze NIRs，then the successive projection algorithm (SPA) and the competitive adaptive reweighted sampling (CARS) were used to extract the characteristic variables. The results showed that LS-SVM model was better than PLS mode. The variable optimization was satisfactory because that the correlation coefficient of prediction (R2p) for hemp seed oil adulterated with soybean oil，peanut oil，sunflower oil and corn oil were 0.9504，0.9058，0.8574 and 0.7673，respectively. In addition，CARS showed better performance than SPA，and the accuracy of LS-SVM-CARS model was the most satisfied with R2p of 0.9821，0.9075，0.9587 and 0.9249，respectively. So，the LS-SVM-CARS model is suitable to discriminate the oil adulteration due to its high accuracy，fewer variables and strong transitivity