摘要
[目的]对比光学特性参数[吸收系数(μa)和约化散射系数(μ′s)]鉴别掺伪茶油的能力,并探索组合提取方式对模型的提升效果。[方法]以植物油为试验材料,制备不同质量分数的掺伪茶油;采用不同的预处理方式对光学特性参数数据进行预处理,提取特征波段后建立随机森林(RF)定性鉴别模型。[结果]经过竞争性自适应重加权算法(CARS)和无信息变量剔除算法(UVE)进行特征提取后,利用μa和μ′s数据建立的模型鉴别准确率分别为95.65%,95.65%和98.55%,97.10%,与CARS特征提取方式相比,组合提取方式(UVE-CARS)使模型的鉴别结果至少可提升1.45%。[结论]利用μa可更快速、准确地实现对不同掺伪种类茶油的鉴别,采用组合提取方式可有效提升模型鉴别能力。
[Objective]In order to compare the ability of optical characteristic parameters[absorption coefficient(μa)and approximate scattering coefficient(μ′s)]to identify adulterated tea oil and to explore the enhancement effect of the combination of extraction methods on the model to achieve a faster and more accurate identification of different kinds of adulterated oils.[Methods]In this study,vegetable oils were used as experimental materials to prepare adulterated tea oils with different mass fractions.Different preprocessing methods were used to preprocess the optical characteristic parameter data,followed by feature band extraction and subsequent establishment of a Random Forest(RF)qualitative identification model.[Results]After CRAS and UVE-CARS feature extraction,the identification accuracies of the models built usingμa andμ′s were 95.65%,95.65%,and 98.55%,97.10%,respectively.The combined extraction method(UVE-CARS)resulted in an improvement of at least 1.45 percentage points in the identification results of the models compared with the CARS feature extraction method.[Conclusion]The identification of different adulterated types of tea oil can be realized more quickly and accurately by usingμa.The combined extraction method can effectively improve the identification ability of the model.
作者
管金伟
李大鹏
龚中良
刘强
蒋涵
GUAN Jinwei;LI Dapeng;GONG Zhongliang;LIU Qiang;JIANG Han(School of Mechanical and Electrical Engineering,Central South University of Forestry and Technology,Changsha,Hunan 410004,China)
出处
《食品与机械》
CSCD
北大核心
2024年第7期30-36,共7页
Food and Machinery
基金
湖南省重点研发计划项目(编号:2022NK2048)
湖南省科学研究重点项目(编号:22A0187)。
关键词
光学特性参数
茶油鉴伪
定性模型
随机森林
optical characteristic parameters
tea oil authentication
qualitative modeling
random forests
作者简介
管金伟,男,中南林业科技大学在读硕士研究生;通信作者:李大鹏(1983-),男,中南林业科技大学讲师,博士。E-mail:dapengli@csuft.edu.cn。
