基于简单而强大神经网络法分析液态奶中的三聚氰胺(IJEME-V1-N3-4)

I.J.EducationandManagementEngineering2011,3,24-30PublishedOnlineSeptember2011inMECS()DOI:10.5815/ijeme.2011.03.04Availableonlineat(2,4,6-triamino-1,3,5-triazine)isanitrogen-richchemicalimplicatedinthepetandhumanfoodrecallsandintheglobalfoodsafetyscaresinvolvingmilkproducts.Duetotheserioushealthconcernsassociatedwithmelamineconsumptionandtheextensivescopeofaffectedproducts,rapidandsensitivemethodstodetectmelamine‟spresenceareessential.Weproposetheuseofspectroscopydata–producedbynear-infrared(near-IR/NIR)andmid-infrared(mid-IR/MIR)spectroscopies,inparticular–formelaminedetectionincomplexdairymatrixes.Itwasfoundthatinfraredspectroscopyisaneffectivetooltodetectmelamineinliquidmilk.Thelimitofdetection(LOD)below1ppm(0.76±0.11ppm)canbereachedifacorrectspectrumpre-processing(pre-treatment)techniqueandacorrectmultivariate(MDA)algorithm:partialleastsquaresregression(PLS),polynomialPLS(Poly-PLS),orartificialneuralnetwork(ANN)–isusedforspectrumanalysis.TherelationshipbetweenMIR/NIRspectrumofmilkproductandmelaminecontentisnon-linear.So,non-linearregressionmethodsareneededtocorrectlypredictthetriazine-derivativecontent.Itcanbeconcludedthatmid-andnear-infraredspectroscopycanberegardedasaquick,sensitive,robust,andlow-costmethodforliquidmilkanalysis.Thetechniquecanbeappliedfortheautomationofmilkanalysis.IndexTerms:food;liquidmilk;partialleastsquaresregression(PLS);artificialneuralnetwork(ANN)©2011PublishedbyMECSPublisher.Selectionand/orpeerreviewunderresponsibilityoftheInternationalConferenceonE-BusinessSystemandEducationTechnology1.IntroductionMelamine(2,4,6-triamino-1,3,5-triazine)isanitrogen-richchemicalimplicatedinthepetandhumanfoodrecallsin2007[1,2]andintheglobalfoodsafetyscaresin2008involvingmilkproducts[3-5].Inthosefoodsafetyincidents,melaminewasintentionallyaddedtofoodsandanimalfeedtoboosttheproteincontent[6,7].Adrivingforcefortheadulterationofafoodproductwithmelamineisthatitshighnitrogencontent,whichincreasestheapparentproteincontentmeasuredbystandardproteinanalysistests,suchasKjeldahlorDumasones[5].NotethattheKjeldahlanalyticalmethodisamethodforthequantitativedeterminationofnitrogenonly(nottrueproteincontent)[8].Correspondingauthor:E-mailaddress:[e-mail]idMilkbySimpleandRobustNeuralNetworkBasedMethod25Inlate2008,traceamountofmelamineweredetectedinUS-madeinfantformulaproducts[9,10].Therecallsinvolvingpetfoodandmilkproducts,contaminatedwithmelamine,havecreatedawidespreadfoodsafetyscare[5,8].Todaymelaminecontaminationhasbeenreportedinavarietyoffoodproductssuchasmilk(liquidorpowder),infantformula,frozenyogurt,petfood,biscuits,candy,andcoffeedrinks[11].The2008incidentofcontaminationofmilkwithmelamineinChinalikelycaused300,000casesofrenalcomplicationsinchildren,andatleast6childdeaths,directlyresultingfromconsumptionoftaintedproduct[5,12].Toxiceffectsassociatedwithmelamineconsumptionoccuronlyfollowinghighdoses[13].Theoral50%lethaldose(LD50)hasbeenreportedasapproximately3gkg-1ofbodyweight[14].Itisthoughtthatsimultaneousingestionofmelamineandoneofitsanalogues,cyanuricacid,mayresultintheformationofcrystalsinthekidney[13],aswasthecaseinthepetfoodincident[15].Weproposetheuseofspectroscopydata–producedbynearinfrared(near-IRorNIR)andmediuminfrared(mid-IRorMIR)spectroscopies,inparticular–formelaminedetectionincomplexmatrixes.NearinfraredspectroscopyincombinationwithMDAmethodshasbeenappliedtomelaminedetectioninanumberofanalyticalstudies[16-35].In2009,Maueretal.[8]evaluatednear-andmid-infraredspectroscopymethods(NIR,FTIR-ATR,FTIR-DRIFT)forthedetectionandquantificationofmelamineininfantformulapowder.Partialleast-squares(PLS)modelswereestablishedforcorrelatingspectraldatatomelamineconcentration:R20.99,RMSECV≤0.9,andresidualpredictiondeviation(RPD)above12.Factorizationanalysisofspectrawasabletodifferentiateunadulteratedinfantformulapowderfromsamplescontaining1ppmmelaminewithnomisclassifications,aconfidencelevelof99.99%,andselectivity2.ItwasstatedthatNIRandMIRmethodsenablerapiddetectionof1ppmmelamineininfantformulapowder[8].Luetal.[36]establishedanovelandrapidmethodfordetectingpuremelamineinmilkpowderusingnearinfrared(NIR)spectroscopybasedonleastsquares-supportvectormachine(LS-SVM).Partialleastsquarediscriminantanalysis(PLS-DA)wasusedfortheextractionofprincipalcomponents(PCs).ThescoresofthefirsttwoPCshavebeenappliedasinputstoLS-SVM.ComparedtoPLS-DA,theperformanceofLS-SVMwasbetter,withhigherclassificationaccuracy,both100%forthetrainingandtestingset.Thedetectionlimitwaslowerthan1ppm.Basedontheresults,itwasconcludedthatNIRspectroscopycombinedwithLS-SVMcouldbeusedasarapidandaccuratemethodfordetectingpuremelamineinmilkpowder.Inthispaper,wetrytoreachtwogoals:(i)toestablishaquick,sensitive(LOD1ppm),reliable,androbustmethodformelaminedetectioninliquidmilkbasedonNIRandMIRspectroscopymethods;(ii)tocomparedifferentm