Electronic nose and data analysis for detection of maize oil adulteration in sesame oil
An!°electronicnose!±hasbeenusedforthedetectionofadulterationsofsesameoil.Thesystem,comprising10metaloxidesemiconductsensors,wasusedtogenerateapatternofthevolatilecompoundspresentinthesamples.Priortodifferentsupervisedpatternrecognitiontreatments,featureextractiontechniqueswereemployedtochooseasetofoptimaldiscriminantvariables.Principalcomponentanalysis(PCA),Fisherlineartransformation(FLT),stepwiselineardiscriminantanalysis(Step-LDA),selectionbyFisherweights(SFW)wereused,respectively.Andthen,lineardiscriminantanalysis(LDA),probabilisticneuralnetworks(PNN),backpropagationneuralnetworks(BPNN)andgeneralregressionneuralnetwork(GRNN)wereappliedaspatternrecognitiontechniquesfortheelectronicnose.AsforLDAandPNN,FLTwasthemosteffectivefeatureextractionmethod,whileStep-LDAwasthemosteffectivewayforBPNNandFLTwasmoresuitableforGRNN.Withonlyonesamplemisclassi?edinourexperiment,LDAismorepowerfulthanPNN.ExcellentresultswereobtainedinthepredictionofpercentageofadulterationinsesameoilbyBPNNandGRNN.Aftertrainingforsometime,BPNNcouldpredicttheadulterationquantitativelymorepreciselythanGRNN,whereaswithFLTasitsfeatureextractionmethodandwithoutiterativetraining,GRNNcouldalsoyieldratheracceptableresults.?2006ElsevierB.V.Allrightsreserved.