基于Elastic Net特征变量选择的黄山毛峰茶等级评价  

Evaluation of Huangshan Maofeng tea grades based on feature variable selection using Elastic Net

作　　者：潘天红 李鱼强[2] 陈琦 陈山 Pan Tianhong;Li Yuqiang;Chen Qi;Chen Shan(School of Electrical Engineering and Automation,Anhui University,Hefei 230061,China;School of Electrical and Information Engineering,Jiangsu University,Zhenjiang 212013,China;Huangshan Customs Research Center for Tea Quality and Safety,Huangshan 245000,China)

机构地区：[1]安徽大学电气工程与自动化学院,合肥230061 [2]江苏大学电气信息工程学院,镇江212013 [3]黄山海关茶叶质量安全研究中心,黄山245000

出　　处：《农业工程学报》2020年第13期264-271,共8页Transactions of the Chinese Society of Agricultural Engineering

基　　金：国家重点研发计划(2017YFF0211301);安徽省高校协同创新项目(GXXT-2019-012)。

摘　　要：为简化茶叶化学检测分析过程,实现茶叶高精度等级评价,该研究以黄山毛峰茶为研究对象,结合茶叶中茶多酚、儿茶素、咖啡碱、没食子酸及氨基酸成分检测,提出基于Elastic Net特征变量选择的茶叶等级评价方法,建立基于特征成分的黄山毛峰茶等级评价模型。试验选取6个不同等级共96个黄山毛峰茶叶样品,并分析了全部样品的19个成分,通过ElasticNet选取了9个特征成分(没食子酸、表儿茶素没食子酸酯、儿茶素、表儿茶素、没食子酸儿茶素没食子酸酯、表没食子儿茶素、谷氨酸、精氨酸和儿茶素苦涩味指数)建立等级评价模型,并与主成分分析(Principal Components Analysis,PCA)进行对比。100次蒙特卡罗试验结果表明,相比于PCA预测集准确率平均值为70.79%,基于Elastic Net特征变量选择的黄山毛峰茶等级评价准确率更高为78.72%。在此基础上,构建Elastic Net特征变量雷达图,实现黄山毛峰茶等级多变量综合评价可视化。研究结果表明所提方法可有效选择茶叶特征成分,提高黄山毛峰茶等级评价准确率,为茶叶高精度等级评价提供参考。Huangshan Maofeng tea has become one of the most famous Chinese tea due to its amazing orchid fragrance and fresh,sweet taste.However,different quality grades of Huangshan Maofeng tea vary greatly in price.The quality evaluation of tea has posed a great challenge in the tea market.The quality grades of variant tea are also related to the different microelements and concentrations.Traditional sensory evaluation methods cannot achieve fast and accurate discrimination,particularly depending on the manual experience.Alternatively,the chemical analysis can serve as an essential method for the quality evaluation of tea.But the chemical analysis for all microelements was confined to its complexity and time-consuming in a large-scale production under gradually refined detection standards with the fast expansion of tea market.Previous studies reveal that the samples collected from the same production or origin places have the similar microelement compositions and concentrations,indicating that the variation of tea grades depends only on a few types of microelements.Therefore,it is reasonable to select the typical microelements for the distinguishing performance,thereby to optimize the traditional chemical analysis.In this work,a new method was proposed based on the feature extraction using the Elastic Net,in order to simplify the procedure of conventional chemical analysis,while to improve the grade evaluation.First,96 samples of Huangshan Maofeng tea were collected from three original places(Fuxi,Yangcun,and Xintian village)with 6 quality grades(advance 1-3 grades,and 1-3 grades)using the traditional manual process.The chemical analysis was used to analyze the types and contents of 19 microelements.Second,a cross-validation method was used to determine the optimal parameters in the Elastic Net,and 9 feature microelements(Gallic Acid,Epicatechin Gallate,Catechin,Epicatechin,Gallocatechin Gallate,Epigallocatechin,Glutamate,Arginine and catechins bitterness index)were selected when the cost function was minimized.Third,the

关 键 词：模型 品质控制 Elastic Net 特征变量选择 黄山毛峰茶 等级评价 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]