同步荧光光谱技术结合支持向量机对掺杂牛奶智能判别研究  

Research on the Synchronous Fluorescence Spectroscopy Combined With Support Vector Machines for Intelligent Discrimination of Milk Adulteration

作　　者：张微微 璩怡 王强 吕日琴 顾海洋 邵娟 孙艳辉 ZHANG Wei-wei;QU Yi;WANG Qiang;LÜRi-qin;GU Hai-yang;SHAO Juan;SUN Yan-hui(Department of Bioengineering,School of Biological Science and Food Engineering,Chuzhou University,Chuzhou 239000,China)

机构地区：[1]滁州学院生物与食品工程学院生物工程系,安徽滁州239000

出　　处：《光谱学与光谱分析》2024年第9期2428-2433,共6页Spectroscopy and Spectral Analysis

基　　金：国家自然科学基金项目(31701685);安徽省教育厅重大基金项目(2022AH040154);滁州市科技计划项目(2021ZD025);滁州学院博士后基金项目(2020BSH002);校级重点研究项目(2022XJZD22)资助。

摘　　要：牛奶因其丰富的营养成分和易消化吸收的特点,受到消费者的青睐。牛奶掺杂行为的产生使得牛奶制品质量备受关注,快速、便捷地鉴别乳品质量对于乳制品行业经济的健康发展具有重要意义。利用同步荧光光谱对掺杂牛奶进行检测,寻求一种高效判别掺杂乳品方法。采用分子荧光分光光度计测定激发波长(Ex)为220~600nm,激发-发射间隔波长(Δλ)为10~180nm的纯牛奶、复原乳粉(全脂、脱脂)及其掺杂样本的牛奶样品的三维荧光光谱数据,利用平行因子分析方法(PARAFAC)降维获取特征光谱,通过支持向量机建立了掺杂牛奶的判别模型。结果所有乳品样品在激发波长为225~300nm范围内都有一个特征荧光峰,荧光峰在280nm附近,为色氨酸类物质,但该处纯牛奶荧光强度明显高于两种乳粉,且脱脂乳粉要强于全脂乳粉,这说明牛奶的主要成分都一样,发光基团一致,但由于处理方式不一致,使得其浓度存在差异。两种复原乳粉在350~400,450~500nm之外存在荧光物质,主要为维生素A和类胡萝卜素,且脱脂乳粉比全脂乳粉对应区域荧光强度要强,主要在于脂肪物质散射使得荧光强度增强。为更好获取乳品样本特征,通过PARAFAC对三维数据进行降维之后,显示当组分数为6,Δλ为40nm时载荷值最大,该处样本信息差异显著。提取Δλ为40nm特征波长和掺杂乳品品类值作为输入数据的支持向量机(SVM)分类器,采用了遗传算法(Ga-SVM)、粒子群优化算法(Pso-SVM)和网格搜索算法(Grid-SVM)三种SVM算法对掺杂牛奶进行分类识别。结果显示Grid-SVM模式交叉验证(CV)准确率为98.91%,其训练集和测试集的分类准确率均为100.00%,且模型运行时间仅6.724s,显著优于另两种分类器。结果表明荧光光谱与PARAFAC-SVM方法相结合,是一种简单且高效判别掺杂牛奶的方法。Milk is favored due to its high nutritional value and consumption rate.Authenticity is a common concern for value assessment.Recently,non-invasive and rapid identification methods have been preferred for the dairy industry.This work proposed a quick method using synchronous fluorescence(SF)spectroscopy and a support vector machine(SVM)for the identification of raw milk.With this aim,SF spectra of milk were recorded between 220and 600nm excitation range withΔλof 10to 180nm,in steps of 10nm.All the milk showed the same fluorescence excitation at band position 280nm,which corresponded to tryptophan.However,the fluorescence intensity of pure milk at this location was significantly higher than that of the two types of milk powder,and skimmed milk powder was stronger than whole milk powder.It indicated that the same main components were in milk.However,there were differences in their concentrations by different treatment methods.Two types of reconstituted formula milk were differentiated based on intensity variations at wavelengths 350~400and 450~500nm.The excitation at these wavelength positions corresponds to vitamin A and carotenoids.At these bands,the skimmed milk powder had a stronger fluorescence intensity in the corresponding region than whole milk powder,mainly due to the scattering of fatty substances,which enhanced the fluorescence intensity.Parallel factor analysis(PARAFAC)was found to reduce threedimensional SF spectroscopy to two-dimensional data,resulting in a better understanding of the characteristics of dairy products.When the suitable components were6,the maximum load value was atΔλwith 40nm,where the difference in sample information was more significant.Then,theΔλwith 40nm and the value of contaminated dairy products as input data were used to classify and identify adulterated milk for the support vector machine(SVM)classifier.The three SVM methods were the genetic algorithm for support vector machine(Ga-SVM),particle swarm optimization support vector machine(Pso-SVM),and grid search algorithm(

关 键 词：荧光光谱 平行因子分析法 掺杂牛奶 支持向量机 

分 类 号：O433.4[机械工程—光学工程]