二维相关近红外光谱的植物油鉴别方法

Research on Vegetable Oils Classification Based on Two-Dimensional Correlation Near-Infrared Spectroscopy

植物油分类鉴别方法研究在食品安全与质量监管中具有重要的研究意义及应用价值。近红外光谱方法可实现复杂成分定性与定量分析,具有无损、快速的优势,在植物油鉴别分类方面应用广泛。研究了基于二维相关近红外光谱的植物油分类鉴别方法:利用傅里叶变换近红外光谱仪采集不同种类植物油的动态光谱,对其进行二维相关分析,得到二维相关同步谱,同步谱的主对角元素即其自相关谱,利用主成分分析提取自相关谱的主成分,最后基于欧氏距离实现植物油种类鉴别。利用二维同步谱的自相关谱主成分之间的欧氏距离实现常见植物油分类判别,提高了分类准确性和算法效率。二维相关分析以正己烷浓度为扰动因素,采集不同浓度扰动下植物油动态光谱,选择不同种类植物油近红外吸收光谱差异最大的6001~6063 cm^-1波长范围计算植物油样本的二维相关谱。二维相关分析的计算过程相对于原始谱而言,其实质就是提取不同品种的植物油随扰动因素变化的特征信息。而这些特征信息体现在大量的数据点中(二维相关同步谱矩阵),因此需要进一步降低变量维度。二维同步谱的对角线元素,即其自相关谱,代表了不同波长处光谱强度随扰动因素变化的程度,这里用自相关谱来代替二维同步谱,大大减少了变量数量。为了进一步降低数据维度,对各种类植物油的自相关谱进行主成分分析,将自相关谱主成分作为分类模型参数。通过计算各种类植物油自相关谱主成分之间的欧式距离,在择近原则基础上实现不同种类植物油鉴别。实验结果表明,二维相关近红外光谱与其特征提取方法相结合可以提高植物油分类准确度,基于自相关谱主成分之间欧式距离的分类方法也为食用油鉴别应用以及自动化处理提供了有效手段。

The research on classifying different kinds of vegetable oil is very important for food safety and quality supervision.The near-infrared spectroscopy could achieve a qualitative and quantitative analysis of samples with complex components.It has been widely applied to classifying different kinds of vegetable oil.The classification method based on two-dimensional correlation near-infrared spectroscopy is applied to recognize typical vegetable oil in this research.Two-dimensional correlation analysis was carried out with the concentration of n-hexane as the disturbance factor.Then the two-dimensional correlation analysis is calculated within the range of 6001~6063 cm-1,in which the absorption features of different kinds of vegetable oils is obvious.And these feature information is reflected in a large number of data points(two-dimensional correlation synchronization spectrum matrix),so further extraction is needed to reduce the variable dimension.The diagonal elements of the two-dimensional synchro spectrum,that is,its autocorrelation spectrum,is always positive,representing the extent to which the spectral intensity varies with the disturbance factor at different wavelengths.Taking advantage of features extracting and data dimensions reduction,the principal component analysis is adapted to extract the feature of the autocorrelation spectrum.The Euclidean distance of principal components is calculated to classify different types of vegetable oil.The experimental results indicated that the proposed method is a benefit for automatic recognition and classification of typical vegetable oil.The PCA algorithm can effectively improve the recognition efficiency and robustness of the model.It provides a new concept for the analysis and processing of food quality sensing spectroscopy.