摘要
采用电子舌技术,针对国内白酒市场上几种常见类型的掺假白酒,对14个白酒样品的酸味、苦味、涩味、苦味回味、涩味回味、鲜味、丰富性、咸味共8个指标进行分析。测定结果显示,掺假白酒与原酒、成品酒在苦味、涩味、苦味回味和鲜味、丰富性等指标之间存在一定程度的差异,表明这些指标在白酒掺假鉴别方面具有应用价值。将14个白酒样品按照原酒、成品酒和掺假白酒分为3类,结合酸味等8个指标的电子舌试验数据,采用机器学习中KNN、决策树2种分类器进行分析,通过交叉验证比较两种分类器的准确率,Python分析结果显示,KNN、决策树等模型的准确率分别为0.9000和0.8667,表明KNN的性能相对更高。综合比较2种分类器准确率,选择KNN作为最终分类器,该模型预测准确率为100%。
The electronic tongue technology was used to target several common types of adulteration of Chinese liquor in the domestic market,starting from 8 indicators such as sourness,bitterness,astringency,aftertaste-B,aftertaste-A,umami,richness and saltiness.About 14 Chinese liquor samples were analyzed,and the results showed that there was a certain degree of difference in the bitterness,astringency,aftertaste-B and umami,richness and other indicators of adulterate liquor and original liquor and bottled liquor,indicating that these indexes had certain application value in the identification of liquor adulteration.At the same time,14 Chinese liquor samples were divided into 3 categories according to the original liquor,bottled liquor and adulterate liquor.Combined with the electronic tongue experimental data of 8 indicators such as sourness,two classifiers such as KNN and decision tree in machine learning were used for analysis,and compared the accuracy of the two classifiers through cross-validation.The results of Python analysis showed that the accuracy rates of KNN and decision tree were 0.9000 and 0.8667,respectively,which showed that the performance of KNN was relatively higher.After comprehensive comparison,KNN was selected as the final classifier,and the accuracy of the model prediction reached 100%.
作者
程铁辕
夏于林
张莹
CHENG Tieyuan;XIA Yulin;ZHANG Ying(The State Key Lab of Liquor Products Test of Chengdu Customs Technology Center Yibin Branch,Yibin 644000)
出处
《食品工业》
CAS
2021年第5期288-291,共4页
The Food Industry
基金
四川省重点研发计划(重大科技专项)项目(编号:2018SZ0360)。
关键词
电子舌技术
白酒
鉴别
KNN
决策树
electronic tongue technology
liquor
identification
KNN
decision tree
