
基于电子鼻的广式焙烤食品质量评判方法研究.pdf
88页分类号:UDC:密级:学校代号:学号:广东工业大学硕士学位论文(工学硕士)118452110803190基于电子鼻的广式焙烤食品质量评判方法研究伍世元指导教师姓名、职称:骆德沤教授企业导师姓名、职称:无专业或领域名称:测这让量技本及仪器学生所属学院:信息工程堂院.论文答辩日期:2Qll笙鱼且鱼目]ADissertationtedtoGuangdongUniversityofTechnologyfortheDegreeofMasterofEngineeringScienceQualityuationofCantonese·-styleBakedGoodsBasedonElectronicNoseMasterCandidate:WuShiyuanSupervisor:Prof.LuoDehanJune2011SchoolofInationEngineeringGuangdongUniversityofTechnologyGuangzhou,Guangdong,P.R.China,510006摘要摘要广式焙烤食品是指珠江流域及南部沿海地区所制作的面点,在我国已享有盛名由于该地区的气候环境、广式焙烤食品易氧化酸败的特点、检测手段的陈旧以及假冒伪劣产品屡屡出现,广式焙烤食品的质量安全问题变得尤其重要。
在对食品质量评判方法的国内外现状和发展趋势进行分析的基础上,本文探索利用电子鼻采集广式焙烤食品的气味信息,通过适当的数据处理方法实现对不同类别和不同存储期的焙烤食品的质量评判研究思路是:选择广式杏仁饼作为研究对象,利用PEN3电子鼻获取不同类别和不同存储期杏仁饼的气味信号;通过传感器信号的基线处理和归一化变换得到原始数据集合;选择能够代表原始数据各个方面的特征组成一个完备的特征集合,通过鉴别结果的反馈对特征集合进行优化,得到一个使判别分类最优的特征子集;用主成分分析方法(PCA)对特征子集数据进行进一步的降维处理:接着采用Fisher线性判别分析(LDA)、马氏距离判别分析以及带动量项的自适应学习率的三层BP神经网络得出判别结果;把这些图形、指标、结论和理化指标(过氧化值、酸价)进行对比,进行区别和联系的分析,实现对杏仁饼的类别和存储期评判选择了6种不同类别的杏仁饼、12个批次的不同存储期的夹肉杏仁饼和12个批次的不同存储期的粒粒杏仁饼作为研究对象其结论如下:(1)不同类别的杏仁饼的鉴别采取以下两种方法:PCA+LDA的方法和PCA+带动量的自适应学习率的三层BP神经网络的方法通过响应曲线图和雷达图分类效果很差;而基于特征子集的PCA方法分类效果有了很大的改善,但是分类面还是靠的很近;PCA+LDA的分类图能够做到类内集中,类间分散的良好效果;通过对比,PCA+BP神经网络的方法(正确率100%)优于PCA+LDA的方法(正确率96%)。
2)不同存储期的杏仁饼的质量评判首先获得不同存储期的夹肉杏仁饼和粒粒杏仁饼的理化指标作为专家数据数据分析采用PCA+LDA的方法,并和理化指标作对比利用响应曲线图和雷达图可以得到存储期不同的杏仁饼的大致变化趋势;通过均值和方差特征图谱得到比较满广东工业大学硕士学位论文意的结果,并根据第二个传感器的相应值给出了验证变质与否的两个指标;而通过特征子集下的PCA图谱可以看出存储期的详细变化趋势,但是内间距靠的很近:通过特征子集下的PCA+LDA的分类图谱能够清晰的区别过期与不过期的杏仁饼;最后通过马氏距离判别分析,把数据分为过期与不过期的两类,对不同存储期的夹肉杏仁饼和粒粒杏仁饼得到正确率96%和97%的判别率AbsttactAbstractCantonese-stylebakedfoodisakindofpastryproductioninthePearlRiverValleyandsoutherncoastalareasofChina.Itiswidelyknownathomeandabroad.Astheregion’Sclimateandenvironment,Cantonese-stylebakedfoodeasytooxidativeranciditytheoldmeansofqualitydetection,andthefakeandshoddyproducts,thequalityandsafetyofCantonese-stylebakedfoodbecomeespeciallyimportant.BasedonthedetailedanalysisofthedevelopmenttrendfortheidentifyclassificationandqualitycontrolresearchofCantonese-stylebakedfoodathomeandabroad,thispaperisaimedtouseanelectronicnosetoidentifyandassessthequalityofCantonese-stylebakedfoodusingthecomprehensiveinationoftheCantonese-stylebakedgoodsvolatilecomponentsthroughappropriatedataprocessing.Research:firstlyselectCantonese-stylealmondcakeastheresearchobjcot,anduseelectronicnosePEN3togetdifferenttypesandstorageperiodsoftheodorination;Thengettheoriginaldatasetbythebaselineprocessingandnormalizationtransofsensorsignal;Choosealltherepresentativecharacteristicsofodorinationtoconstituteacompletefeatureset;thenobtainthemosteffectivefeaturesubsetbyoptimizingthefeedbackoftheextractfeatures.Usingprincipalcomponentanalysis(PCA)onthefeaturesubsetforfurtherreduceddimensionality;FollowedbyFisherlineardiscriminantanalysis(LDA),Mahalanobisdistancediscriminantanalysisandtheamountofitemsdrivingthethree—layeradaptivelearningrateofBPneuralnetworkidentificationresultsobtained;Finallythesegraphics,index,conclusionsAndthephysicalandchemicalindicators(peroxidue,acidvalue)werecomparedfordifferencesandcontactanalysis.Accordingtotheresearchtarget,choosingthe6differenttypesofalmondcake,12batchesofdifferentstorageperiodsalmondcakewithmeatand12differentbatchesofdifferentstorageperiodsLi-lialmondcakeforthestudy.Theconclusionsareasfollows:(1)ClassificationandidentificationalmondcakewithdifferenttypesIII广东工业大学硕士学位论文Takethefollowingtwos:PCA+LDAandPCA+theamountofitemsdrivingthethree—layeradaptivelearningrateofBPneuralnetwork.Thereisaverypoorclassificationthroughresponsecurvesandpolarplots;andPCAbasedonfeaturesubsetclassificationperancehasbeengreatlyimproved,butclassificationsurfacestillrelyonthecloseside;PCA+LDAclassificationmapsCanbeconcentratedwithintheclass,andbetween·classspreadclearly;Bycontrast,PCA+BPneuralnetwork(accuracyrate100%)issuperiortoPCA+LDA(accuracyrate96%).(2)JudgethequalityofalmondcakewithdifferentstorageperiodsFirstlyToobtainthephysicalandchemicalindicatorsasanexpertdataofalmondcakewithmeatandLi—lialmondcakewithdifferentstorageperiods.DataanalysissusingPCA+LDA,andphysicalandchemicalindicatorsforcomparison.UsingresponsecurvesandpolarplotsCangetthetrendofalmondcakewithdifferentstorageperiods.Characteristicspectrumbythemeanandvariancearequitesatisfactoryandinaccordancewiththecorrespondingvaluesofthesecondsensorisgiventoverifywhetherornotthetwometamorphicindicators;AndthroughthePCAmapunderfeaturesubsetCallgetthetrendsoftransationwithdifferentstorageperiodsclearlybutthespacingofcategoriesisclosely.ThroughthePCA+LDAclassificationmapunderfeaturesubsetCandistinguishexpiredandexpiringalmondcake.Finallyusingmahalanobisdistancediscriminantanalysis,andthedataisdividedintotwocategoriesofexpiredandexpiringonthedifferentstorageperiods,almondcakewithmeatandLi—lialmondcaketogettherightrateof96%and97%classificationrate.Theresultsshowthatit’SpossibletoclassifyandidentifythequalityofCantonese-stylebakedgoodsaccuratelybyaportableelectronicnosePEN3.Keywords:electronicnose;qualityofCantonese-stylebakedgoods;featureextractionandselection;back-propagationneuralnetwork;principa。
