Research On Discriminant Analysis Of Milk Adulteration By Infrared Spectroscopy

Since the beginning of the 21st century, the issue of food safety is becoming a global concern. The methods currently used for food safety detection are usually with high cost and complex operation. Moreover, these methods are usually designed for the specific components. Thus, there is an important significance for developing a rapid, cost-effective, and widely available method for food safety detection. In this research work, infrared spectroscopy techniques and pattern recognition were applied to study the qualitative discriminant analysis method for milk adulteration detection.Starting from the physical structure and composition of milk, the spectral characteristics of milk in mid-infrared (MIR) and near-infrared (NIR) regions were analyzed. Then, samples were doped with one of three adulterants, urea, glucose and melamine. The MIR and NIR spectra of each type of adulterated samples were measured and the effect on the typical spectral characteristics by the adulterants was also analyzed. The background correction method, secondary derivative spectra method and two-dimensional correlation spectroscopy method were respectively used for spectral analysis. The results showed that, in the mid-infrared spectral regions, the above three methods can be used to preliminarily discriminate whether the milk was adulterated according to the comparison with spectral characteristics of the samples, such as the location and intensity of absorption peeks.Chemometrics methods combined with pattern recognition were used for qualitative discriminant analysis of milk adulteration. The samples were adulterated milk with one of the three adulterants, urea, glucose and melamine with different concentrations. Soft independent modeling of class analogy (SIMCA), Bayes discriminant analysis, and partial least squares discriminant analysis (PLSDA) were respectively used to construct discriminant models and the models were compared with each other. The results showed that the PLSDA model was the best.The optimization method of the model was studied. The effect of different spectral pretreatment methods and different band selection on the results of discrimination was analyzed. Through comprehensive comparison of modeling results, the Savitzky-Golay convolution smoothing through 5 points with combination of standard normal variate was chosen as the best spectral pretreatment method and the PLSDA model had the best results in near-infrared 5896-4000 cm-1 and the mid-infrared 1800-704 cm-1. PLSDA models were constructed respectively for each type of adulterated sample sets (urea, melamine and glucose) and then a PLSDA model was also constructed by all the three types of adulterated sample sets. The results showed that, the model achieved 93.2% and 92.6% discriminant accuracy respectively in NIR and MIR regions, in the classification of different adulterated and unadulterated milk samples. Thus, it can be concluded that infrared spectroscopy and PLSDA can be used to identify whether the milk has been adulterated or not and the type of adulterant used.