植物油市场中出售的芝麻油、玉米油和花生油有多种品牌, 不同品牌间价格差距较大, 且存在假冒的现象, 利用荧光光谱技术可以无损地鉴别购买油种是否为标签所标种类。主成分分析方法及平行因子方法可对这3种油种进行人工分类, 但其存在类间距离相比于类内距离过小的不足, 在结合传统的聚类分析方法时, 会造成误分类现象。本文以提高类间距离、达到正确聚类为目标, 经过比较分析, 选择均值、标准差、光谱重心坐标、二阶混合中心距、相关系数、等价椭圆二倍倾角正切值、在重心激发波长处的发射光谱的偏度系数和峰度系数作为统计参数, 相比于直接使用聚类方法, 芝麻油分类的正确率从92.3%提高到100%, 玉米油分类的正确率从75%提高到100%, 花生油从57.1%提高到100%。用偏最小二乘判别分析方法验证了本文方法的合理性。本文方法可以用于植物油检测仪器的自动分类, 利于市场监管及指导人们日常消费。

There are many brands of sesame oil, corn oil and peanut oil in the vegetable oil market. There is a big difference among the price of different brands, and the phenomenon of counterfeiting is existing. Fluorescence spectroscopy can be used to identify the real label of the oil species. Principal component analysis and parallel factor method can classify manually three species of oils, but there is a problem that the distance between classes is too small compared to the in-class distance, and it is easy to cause misclassification when the traditional clustering analysis method is used. In order to improve the distance between classes to achieve the correct clustering, the mean, standard deviation, gravity coordinates of spectral center, second-order mixed center distance, correlation coefficient, equivalent elliptical double diagonal tangent, the skewness coefficient and the kurtosis coefficient of the emission spectrum at gravity excitation wavelength are selected as statistical parameters after the comparative analysis. Compared to the traditional clustering methods, the accuracy rate of sesame oil classification increased from 92.3% to 100%, and the accuracy of corn oil classification increased from 75% to 100% and the accuracy of the peanut oil classification increased from 57.1% to 100%. Finally, partial least squares discriminant analysis is used to verify the validity of the selected method. The method used here can be used for automatic classification of vegetable oil in detection equipment, which will help regulate the market and guide peoples daily consumption.