本文以2011与2010年间从上海地区收集的牧场鲜奶为样本,通过向原奶中添加 5%一15%浓度的三聚氰胺!尿素!硝酸钱的糊精!淀粉水溶液(其按凯氏定氮法测得的/蛋 白质0含量为3%),构造了一系列接近实际状况的掺假牛奶基于不同的样本集合建立 了各种掺假原奶的近红外模式识别模型,以考察近红外技术识别含有不同伪蛋白及掺假 固形物掺假牛奶的可行性!进一步判断是否可以利用近红外技术判断掺加的伪蛋白与固 形物种类 首先针对样品的近红外测试模式以及预处理方式进行了研究比较,发现将牛奶样品 在40e下超声20min!采用漫反射测试模式进行检测,可以获得高质量的近红外光谱 其次,应用偏最小二乘判别(PLS一DA),K最近邻(KNN),改进与简化的KNN(IS.KNN) 以及支持向量机(SVM)等模式识别方法对真!假奶两类分类结果进行了比较,结果表明 IS一KNN以及SvM方法所建近红外模型的判别效果最好,结果最为稳健 采用这两种方法基于不同样本集合建立的对真!假奶识别的两类分类近红外判别模 型,绝大部分情况下均有高达90%以上的判别正确率基于这两种方法建立的判别掺假 固形物种类的多类近红外分类模型的判别正确率通常高于80%,但判别掺假奶中伪蛋白 品种的多类分类模型的判别正确率只有40%一78.5%进一步分析结果表明所建立近红 外模型的判别正确率与掺假溶液浓度呈正相关关系,近红外模型判别结果与掺假奶中伪 蛋白和掺假固形物品种无关,提示本文所建近红外模型可良好识别掺假奶的主要原因在 于超出近红外检测限的水被加入掺假奶中对近红外光谱产生影响所致 综上,采用15~KNN与SVM方法建立的近红外模型可良好识别掺假溶液浓度在5% 以上的掺假原奶,且模型的正确率随掺假溶液浓度的增高而提高如需进一步确定掺假 奶中伪蛋白种类,可借助试剂盒!色谱等针对某种伪蛋白(如三聚氰胺)的分析方法 关键词:近红外光谱技术;掺假牛奶;IS~KNN;SvM;伪蛋白第11页华东理工大学硕士学位报告 FeasibilitystudyondiseriminatingadulteratedmilkSbynearinfrared sPectroscoPy AbstraCt InthisPaPer,hundredsofraweowmilksamPleswereeollectedfromthepasturesof Shanghaiandneighbourareasin2011and2010.AseriesofadulteratedmilksamPleswere PreParedthroughaddings%一15%adulterationsolutionofdextrinorstarehthateontained melamine,ureaorammoniumnitratetotherawmilks.Andthe-Protein.deteetedbyKjeldahl methodintheadulterationsolutionswas3%.Inordertoinvestigatethefeasibilityof aPPlyingnearinfrared(NIR)technologytodiseriminatingadulteratedmilkaddeddifferent Pseudo一Proteinsandsolidcontents,andfurthermoretojudgeiftheNIRteehniqueean diseriminatetyPeofPseudo一Proteinsandaddedsolidcontents,variousNIRPattern reeognitionmodelswereestablishedbasedonthedifferentsamPlesets. First,themodesoftestingmilksamPlesandthemethodsofPretreatingsamPleswere studiedandeomPared.ItwasindicatedthathighqualitymilkNIRsPeetraeanbeensuredby ultrasoundingmilksamPles20minat40oCbeforecolleetingtheirNIRsPeetrumwithdi而se refieetaneemode. Seeondly,thePattemreeognitionmethodsofPartialleastsquaresdiserimination (PLS一DA),K一nearestneighbour(KNN),imProvedandsimPlifiedKNN(15一KNN)and suPPortveetormaehine(SVM)Wereusedtodiseriminatetrueandadulteratedmilkinthis Paper.Thetwo一elasselassifieationresultsoftrueandfalsemilksbasedonthefourmethods showedthatthe15一KNNandSVMdiseriminatingmethodsalwaysgavebetterandmore robustresults. Fortwo一elasselassifieationoftrueandfalsemilks,theNIRPatternreeognitionmodels establishedby15一KNNandSVMbasedondifferentsamPlesetseanPredietmorethan90% samPlescorreetlyinmosteases.ThePredietionaeeuraeyofmulti一elasselassifieationmodels builtbythetwomethodsfordiseriminatingsPeeiesofadulteratedsolidswasusuallyhigher than80%,however,thatfordiseriminatingtyPeofPseudo一Proteinsaddedtoadulteration milkswasonly40%一78.5%.ThefurtheranalysisshowedthatthePredietionaeeuraeyofthe NIRPattemreeognitionmodelswasPositivelyeorrelatedwiththeeoneentrationofthe adulteratedsolutionsandwasindePendentofthesPeciesofPseudo一Proteinsandadulterated solideontents,ThemainreasonthattheestablishedNIRmodleseanwelldiseriminate adulterationmilksmaybethatWateraddedtomilks15WelloverthedeteetionlimitsofNIR, 50itseffeetonNIRsPeetraofadu]terationmilkscanbereflectedintheNIRmode]s, Insummary,theNIRelassifieationmodelsestablishedby15一KNNandSVMeanwell华东理工大学硕士学位报告第111页 identifyadulteratedmilksamPles,whiehareaddedmorethans%adulterationsoultions.The aecuraeyoftheNIRmodelsinereaseswiththeeoneentrationoftheadulteratedsolutions.The tyPeofPseudo一Proteinsinadulteratedmilkseou]dbedeterminedbykits,ehromatograPhyete aimingtoasPecifieProhibitedPseudoProteinsuehasmelamineifneeessary. Keywords:NIR:Adulteratedmilk;15一KNN:SVM:Pseudo一Protein第IV页华东理工大学硕士学位报告 目录 前言..................................................................................................................1 第1章文献综述.........................................................................................2 1.1近红外技术的基本原理.................................,....................................................,2 1.2牛奶中三聚氰胺的检测方法.....................,.........................................................3 1.2.1ELISA快速检测..,.............................................................................................3 1.2.2荧光光度法..............................................................................................................3 1.2.3表面增强拉曼光谱法..............................................................................................3 1.2.4单扫描极谱法...,................................................................................................4 1.2.5高效毛细管电泳法.,!,............................................................,...................4 1.2.6离子色谱法............................................................................................................,.4 1.2.7高效液相色谱法(HPLC).........................................................................................4 1.2.8液相色谱一质谱联用法(LC一MS)..............................................................,.........5 1.2.9气相色谱一质谱联用法(GC一MS).............................................................................5 1.3近红外技术在食品!牛奶质量检测中的应用.......................,...........................6 1.3.1近红外光谱技术在牛奶及其制品定量分析中的应用..........................................6 1.3.2近红外光谱技术在牛奶及其制品定性分析中的应用..........................................6 1.4近红外技术模式识别方法在相关领域的应用......................,.........................,二9 1.5本文的研究目的与研究思路...................................................................................10 第2章样品的预处理及测试模式的优化选择!光谱预处理的影响...11 2.1样品的预处理..............................................................................................,......H 2.2样品测试模式的选择.................................................................................,.......12 2.3光谱预处理方式对模式识别结果的影响............................................,............13 2.3.1常用光谱预处理方法介绍....................................................................................13 2.3.1.1一阶导[84l...............................................................................,......................13 2.3.1.2标准正态变换(SNV)[84]......................................................................................14 2.3.1.3多元散射校正法(MsC)184]................,.......................................,............14 2.3.2实验部分.....................,.............................:................................,.........巧 2.3.2.1实验仪器及试剂............................................................,..............................巧 2.3.2.2样本来源及制备.........................................................................,.............巧 2.3.2.3实验方法.............................................................................................................巧 2.3.3不同预处理方法对于模式识别结果比较.........!...,..................................16华东理工大学硕士学位报告第V页 2.4小结.........................................................................................................................,.19 第3章多种模式识别方法进行掺假牛奶识别的适应性研究...............20 3.1不同模式识别方法的原理.......................................................................................20 0q`0,勺白-1PLS一