Discrimination of Doped Milk by Two-Dimensional Correlation Infrared Spectroscopy in Combination with Support Vector Machine and Gray Level Co-Occurrence Matrix

Discrimination of doped milk by two-dimensional correlation infrared spectroscopy in combination with support vector machine (SVM) and gray level co-occurrence matrix was carried out on 3 kinds of doped milk samples, i.e., urea-doped, melamine-doped and glucose-doped milk samples. Two-dimensional correlation infrared spectra of doped milk was established with mass concentration as the external disturbance. Second moment of angle, moment of inertia of main diagonal, correlation coefficient, mean and standard deviation of entropy were selected as the image texture features, and SVM models for discrimination were established for the 3 kinds of doped milk samples. It was shown by the results, that when the synchronous spectra were studied, values of accuracy of discrimination for the training sets of the 3 kinds of doped milk samples were 91.7% (for urea-doped milk), 96.7% (for melamine-doped milk) and 91.7% (for glucose-doped milk) and values for prediction sets were 85.0%, 90.0% and 100% respectively. Further tests for urea-doped milk were done by study with combination of synchronous spectra and asynchronous spectra, giving values of accuracy of discrimination for samples of training set of 98.1% and for samples of prediction set of 92.3%. The elevation of accuracy of discrimination in this case was found due to obtaining more information by using the synchronous and asynchronous spectra in combination which was more effective for the discrimination. It was concluded that the proposed method was feasible for discrimination of doped milk.