基于GC-IMS的白酒特征分析及鉴别

Feature Analysis and Identification of Baijiu Based on Gas Chromatography-Ion Migration Spectrometry

探讨一种在未对挥发组分定性、定量的条件下,实现白酒准确鉴别的简便方法。选用两种香型共10种白酒,每种白酒取2个勾调批次的样品。直接原样顶空制样,通过气相色谱-离子迁移谱法(GC-IMS)分析,获得可视化的指纹图谱,并进一步提取白酒主要特征建立Fisher线性判别模型。从白酒样品中共检出134种挥发组分。指纹图谱可直观反映白酒样品中微量挥发组分的含量差异和勾调的批次稳定性。在未完全剖析和定量所有组分前提下,采用逐步回归分析分别对基于全部134种挥发组分、相关性分析选择的39种特征组分以及主成分分析提取的6个主成分建立的3个Fisher线性判别模型进行优化。优化后的相关性分析模型(优化为21种组分特征)和主成分分析模型对训练集的分类正确率和留一法交叉验证分类正确率均为100%,对测试集样品的判别正确率均为95%。由全部特征经逐步回归分析优化的模型,共提取24个组分特征,建立的Fisher线性判别模型效果最优。其对训练集分类正确率为100%,留一法交叉验证分类正确率为100%,对测试集酒样预测正确率达100%。GC-IMS方法可为白酒生产过程中的质量控制、市场监管及白酒原产地保护提供技术支持,结合模式识别方法可对白酒进行准确鉴别。

A simple method for accurate identification of Baijiu was developed without qualitative and quantitative determination of volatile components.A total of 10 kinds of Baijiu with two flavor types were selected,and each of them has2 samples from different blending batches.After directly headspace sample preparation,gas chromatography-ion mobility spectrometry(GC-IMS)was used to obtain the visualized fingerprints,and further extract the main features of Baijiu to establish Fisher linear discriminant models.A total of 134 volatile components were detected in the samples.The fingerprints can directly reflect the content difference of trace volatile components and the batch stability of blending.Under the premise of not fully analyzing and quantifying all components,stepwise regression analysis is used to optimize the three Fisher linear discriminant models which are respectively based on all 134 volatile components,39 characteristic components selected by correlation analysis,and the 6 principal components obtained from principal component analysis.For the optimized correlation analysis model(the component features are reduced to 21)and the principal component analysis model,the classification accuracy rates and leave-one-out cross validation classification accuracy rates for the training set are 100%,and the discrimination accuracy rates for the test set are both 95%.The optimized Fisher linear discriminant model based on all the features has the best result,and 24 component features are extracted.The model’s classification accuracy rate of the training set is 100%,the leave-one-out method cross-validation classification accuracy rate is 100%,and the prediction accuracy rate of the Baijiu samples in the test set reaches 100%.The GC-IMS is significant for the quality control,market supervision and the origin protection of Baijiu,and it can accurately identify Baijiu combining with pattern recognition.