基于一维卷积神经网络结合便携式拉曼光谱特级初榨橄榄油掺假定量分析  

Quantitative analysis of adulteration in extra virgin olive oil based on one-dimensional Convolutional neural network and portable raman spectroscopy

作　　者：张焕俊 戴臻[2] 费洪晓 ZHANG Huanjun;DAI Zhen;FEI Hongxiao(Zhumadian vocational and technical college,Zhumadian,463000,Henan,China;Hunan vocational college of science and technology,School of Software,Changsha,410004,Hunan,China;Central south university,school of computer science and engineering,Changsha,410012,Hunan,China)

机构地区：[1]驻马店职业技术学院,河南驻马店463000 [2]湖南科技职业学院,软件学院,湖南长沙410004 [3]中南大学,计算机学院,湖南长沙410012

出　　处：《光散射学报》2024年第4期436-444,共9页The Journal of Light Scattering

基　　金：湖南省自然科学基金委员2021年科教联合课题(2021JJ60048)。

摘　　要：针对造假手段的不断提升的现状及廉价的橄榄果渣油很可能成为特级初榨橄榄油掺假的潜在原材料等问题。因此,本研究重点围绕深度学习算法辅助非接触式无损伤光谱检测技术量化特级初榨橄榄油的掺假行为。将过期橄榄果渣油和特级初榨橄榄油按不同体积比例混合,从而制备出不同浓度的掺假混合油品。使用785 nm便携式拉曼光谱仪对这些混合油品进行拉曼光谱采集,并结合一维卷积神经网络算法建立掺假量化模型。采用密度泛函理论基于B3LYP/6-31+G(d,p)基组计算亚油酸分子的理论振动光谱,以进一步解析特级初榨橄榄油的拉曼光谱。结果表明,基于具有深度结构的前馈神经网络与785 nm便携式拉曼光谱技术联用的技术方案是定量分析植物油掺假的有力工具,80个混合油品的4000条光谱数据量化模型的决定系数均优于0.97,其中评价模型测试集定量分析的决定系数达到了0.9704,均方根误差小于0.0499。该技术在快速评估特级初榨橄榄油掺假方面具有很好的应用潜力,为规范国内橄榄油市场和维护消费者合法权益提供了一种有益的参考方案。Faced with the continuous improvement of counterfeiting methods,cheap olive pomace oil will probably become a potential raw material for adulterating extra virgin olive oil.Therefore,this study focuses on the deep learning algorithm-assisted non-contact,non-destructive spectral detection technology to quantify the adulteration behavior of extra virgin olive oil.Mix expired olive pomace oil and extra virgin olive oil in different volume proportions to prepare different adulterated,mixed oil concentrations.The 785 nm portable Raman spectrometer was used to collect the Raman spectra of these mixed oils,and the quantitative analysis model of adulteration was established by combining the one-dimensional convolutional neural network algorithm.The density functional theory B3LYP/6-31+G(d,p)basis set was used to calculate the Raman spectrum of linoleic acid molecules to further analyze the Raman spectrum of extra virgin olive oil.The experimental results show that the technical solution based on combining deep structured feedforward neural networks and 785 nm portable Raman spectroscopy technology is a powerful tool for quantitative analysis of plant oil adulteration.The decision coefficients of 4000 spectral data quantitative models from 80 mixed oil products are all better than 0.97,and the decision coefficient of quantitative analysis in the evaluation model test set reaches 0.9704,with a root mean square error less than 0.0499.This technology has great application potential in quickly evaluating the adulteration of extra virgin olive oil,providing a beneficial reference scheme for regulating the domestic olive oil market and safeguarding consumers’legitimate rights and interests.

关 键 词：一维卷积神经网络 便携式拉曼光谱 密度泛函理论 特级初榨橄榄油 橄榄果榨油 掺假量化 

分 类 号：O433.4[机械工程—光学工程]