Classification of adulterated milk with the parameterization of 2D correlation spectroscopy and least squares support vector machines

Abstract

A new discriminant method to classify adulterated milk and pure milk is proposed by combining the parameterization of two-dimensional (2D) correlation spectroscopy with least squares support vector machines (LS-SVM). 120 pure milk samples and 120 milk samples adulterated with melamine, urea and glucose were prepared and their synchronous 2D correlation spectra were calculated. Then 5 statistical parameters, which were mean, variance, standard deviation, skewness and kurtosis, were extracted based on the parameterization theory. Finally, the discriminant model of adulterated milk and pure milk was built combining these statistical parameters with LS-SVM. The ratios of correct classification 96.3% and 90% for calibration set and prediction set, respectively, were obtained. The results show that this method can not only extract effectively feature information of adulterant in milk, but also reduce the input dimension of LS-SVM and computational time demands, and so better fitted to realize discriminant analysis of adulterated milk and pure milk.