Feasibility Study On Discriminant Analysis Of Milk Adulteration Based On Optical Parameters Measurement

In recent years,the issue of food safety is severe and becoming a global problem.At present,the adulteration detection method has the characteristics of high requirements on environment and sample,high cost and complicated operation.Moreover,these methods are usually designed for the specific and known components,which is not conducive to popularization.Therefore,as a rapid,cost-effective,non-polluting method,spectroscopy techniques are widely used for food safety detection.In this paper,near infrared spectroscopy(NIR)techniques and pattern recognition methods are applied to qualitatively discriminate the milk adulteration detection based on optical parameters measurement.(1)The milk samples are prepared with different fat,protein and water contents,and the optical parameters are measured by the double-integrating sphere system and inverse doubling method.Then the influence of the main components on the optical parameters is analyzed.The result shows that the scattering coefficient is sensitive in the range of 1140-1400 nm,and it is highly correlated with the fat content in the range of 1300-1400 nm.The absorption coefficient is sensitive in the range of 1400-1500 nm and the direction of change in the range of 1200-1300 nm is different in 1500-1600 nm.This indicates that the optical parameters contain milk composition information,and they can be used for adulterated milk identification.(2)Different types of milk from Mengniu,Haihe,Yili pure milk and Yili high calcium milk are selected as samples.And based on the absorption coefficient and scattering coefficient,the unsupervised pattern recognition method,k-means is used to classify the milk.And the accuracy is above 97.5%.However,this method does not recognize samples incorporating urea and melamine.(3)Urea(0.05 g-0.6 g)and melamine(0.1 g-0.3 g)are added into the milk as samples.Qualitative discriminant analysis is carried out by Principal Component Analysis and Partial Least Squares Discriminant Analysis,the supervised pattern recognition method,respectively.The result shows that the discrimination efficiency of Partial Least Squares Discriminant Analysis method is better than that of Principal Component Analysis method,and its discriminative accuracy of single dopant discrimination model is more than 70%.In addition,the discriminative accuracy of all samples for the determination of urea,doping melamine and pure milk is 90%,80% and 70% respectively.Furthermore,based on the sensitivity of the absorption coefficient and the scattering coefficient,the Partial Least Squares Discriminant Analysis method is optimized for band selection.The scattering coefficients in the range of 1140-1400 nm and 1500-1600 nm and the absorption coefficients in the range of 1150-1650 nm are selected and the discriminative accuracy of the models is more than 80%.Therefore,based on the optical parameters of milk it can be achieved on adulterated milk identification.