Study On Discrimination Of Freshness Of Goat's Milk And Adulteration Goat's Milk By Near Infrared Spectroscopy Technology

As a food with rich nutrient, goat's milk and goat's milk products have been consumedmore and more by people. But there exist some new quality problems of goat's milk, such asdeterioration goat's milk, adulteration raw milk. They are serious threat to the safety ofChina's dairy industry. So it is very important to develop a quick and accurate technology torapid detection of freshness of goat's milk, adulteration goat's milk for market managementand ensure milk safety.At first, non-destructive freshness assessment of goat's milk during four days of normaltemperature preservation and nine days of cold storage were carried out by means of anFT-NIR spectrometer and a fiber optic probe. After each spectral acquisition, Hierarchicalcluster analysis was also conducted to test similarity between values at different days ofstorage. The freshness parameters such as acidity and pH value were destructively measured.For all milk samples, PCA(principal component analysis) and BP neural network, werecarried out in order to set up models to predict the freshness parameters and to classify milksamples according to the days of storage.Then,424samples of goat's milk and adulteration milk were collected according to theadulteration concentration, these adulteration milk has seven categories which were mixedwith margarine, reconstituted milk, starch, cow's milk, urea, sodium nitrite, separately, we usethese goat's milk and adulteration milk as study object to carried out the study of rapiddetecting of adulteration milk through near-infrared (NIR) spectroscopy combined withchemometrics methods, the aim of this research was to establish NIR discriminant modelbetween goat's milk and six kinds of adulteration milk, NIR classification discriminant modelof six kinds of adulteration milk and quantitative models of the content of adulterationmaterial in adulteration milk.Specific results were as follows:BP neural network was carried out in order to set up models to predict the freshnessparameters (acidity and pH) of goat's milk, As for normal milk, result indicated that R2valueof quantitative calibration model are99.50%,98.52%, separately, RMSECV are1.1374,0.0815, separately; the R2value of validation set are98.87%,89.39%, separately, RMSEP are1.3742,0.1634, separately; As for cold milk, result indicated that R2value of quantitative calibration model are99.55%,97.68, separately, RMSECV are0.1208,0.0432, separately; theR~2value of validation set are88.80%,95.00%, separately, RMSEP are0.5291,0.0740,separately;A near infrared two categories discriminant model between goat's milk and six kinds ofadulterated milk has been build,result indicated the correct distinguishing rate of36unknowntest samples is94.4%. Near-infrared classification discriminant models of six kinds ofadulterated milk were then build by RBF(Radial Basis Function) and MLP (MultilayerPerceptron) neural network. The rate of correct discriminant of the models built by RBF is98.8%and the prediction accuracy is96.6%;The rate of correct discriminant of the modelsbuilt by MLP is99.6%, and the prediction accuracy is99.1%; Finally, quantitative analysismodels of the content of adulteration material in adulteration milk were build by PLS, the R2(The Coefficient of Determination) value of quantitative calibration model of milk adulteratedwith margarine, milk adulterated with reconstituted milk, milk adulterated with starch, milkadulterated with cow's milk, milk adulterated with urea, milk adulterated with sodium nitriteare98.85%,97.06%,96.64%,98.14%,96.16%,91.31%, separately, RMSECV (Root MeanSquare Error of Calibration) are0.333,5.61,1.52,4.37,0.519,0.093; the R2value ofvalidation set are90.989,0.982,0.979,0.964,0.951,0.867, separately.Near infrared spectroscopy was combined with PLS-DA (Partial least squares differenceanalysis), fisher's linear discriminator and MLP artificial neural network to built calibrationmodels, between pure goat's milk and blends adulterated milk. The rate of correctdiscriminant of the three models were93.25%,96.9%,97.5%, separately, and the predictionaccuracy were90.91%,98.9%,100%, separately.All of these suggested that near infrared spectroscopy has good potential to quantitativeand qualitative detect goat's milk rapidly and non-destructively.