紫外光谱结合BP神经网络算法建立食用油掺伪煎炸油的快速鉴定模型  

Establishment of Rapid Identification Model of Edible Oil Adulterated with Frying Oil by Ultraviolet Spectrum Combined with BP Neural Network Algorithm

作　　者：陈林林[1] 吴松遥 王玲 张铭 李昕彤 张海鹏 郝熙 李伟[1] Chen Linlin;Wu Songyao;Wang Ling;Zhang Ming;Li Xintong;Zhang Haipeng;Hao Xi;Li Wei(College of Food Engineering,Harbin University of Commerce,Haerbin 150028)

机构地区：[1]哈尔滨商业大学食品工程学院,哈尔滨150028

出　　处：《中国粮油学报》2024年第6期206-214,共9页Journal of the Chinese Cereals and Oils Association

基　　金：黑龙江省省属高等学校基本科研业务费科研项目(2023-KYYWF-1054);黑龙江省“百千万”工程科技重大专项(2021ZX12B07-1)。

摘　　要：为建立一种快速食用油掺伪煎炸油检测方法,采用紫外光谱法鉴别其掺伪,本研究选取大豆油、玉米油和葵花籽油为代表分别煎炸,在纯油中掺入相应煎炸时间0~6 h及掺假梯度0%~90%的煎炸油制备掺伪油样,进行紫外光谱及二阶导数预处理,经处理后的光谱特征峰与BP(Backpropagation)神经网络算法结合建立食用油掺伪煎炸油模型,对掺入煎炸油类别、煎炸时间和煎炸油含量进行鉴别分析。结果表明二阶导数预处理后掺伪煎炸油的光谱特征峰中大豆油为446、462 nm、玉米油为268、274 nm、葵花籽油为280、288 nm,根据其特征峰位与峰值建立Levenberg–Marquardt算法(LMA)、动量梯度下降法(MGD)及弹性梯度下降法(EGD)掺伪模型识别率分别为98.15%、91.67%、95.52%。To establish a method for rapidly detecting adulteration in frying oils,ultraviolet spectroscopy was employed to distinguish adulterants.In this study,soybean oil,corn oil,and sunflower seed oil were selected as representative oils and were subjected to frying.Adulterated oil samples were prepared by blending pure oils with frying oils for different frying durations(0-6 h)and varying levels of adulteration(0%-90%).The ultraviolet spectra were pre-processed by using a second-derivative transformation,and a model for detecting adulteration in frying oils was established by combining the processed spectral characteristics with the Backpropagation(BP)neural network algorithm.This model was used to analyze the type of adulterant,frying time,and adulterant content.The results indicated that after the second-derivative transformation,the spectral characteristic peaks for adulterated frying oils were as follows:446 nm and 462 nm for soybean oil,268 nm and 274 nm for corn oil,and 280 nm and 288 nm for sunflower seed oil.In accordance with these peak positions and values,the recognition rates of the Levenberg-Marquardt algorithm(LMA),Momentum Gradient Descent(MGD),and Elastic Gradient Descent(EGD)for identifying the adulteration model were 98.15%,91.67%,and 95.52%,respectively.

关 键 词：食用油 煎炸油 紫外光谱 掺伪 BP神经网络算法 

分 类 号：TS225.1[轻工技术与工程—粮食、油脂及植物蛋白工程]