Identification of adulterated milk powder based on convolutional neural network and laser-induced breakdown spectroscopy

Abstract

Nowadays, milk powder adulteration raises lots of attention, as the nonstandard and illegal exogenous protein adulteration in milk powder could cause serious health problems (e.g., malnutrition and digestive function damage) to infants and adults. Therefore, how to detect the exogenous protein adulteration in milk powder becomes a hot research point. Herein, we developed laser-induced breakdown spectroscopy (LIBS) technique for rapid and all-element analysis of milk powder. Further, the traditional machine learning methods and convolutional neural network (CNN) were adopted to realize the accurate identification of various adulterated milk powders. Twenty-five pieces of milk powder mixed with four different types of exogenous proteins were prepared for the experiment. Four typical machine learning methods, linear discriminant analysis (LDA), k-nearest neighbor (KNN), random forest (RF), and support vector machines (SVM), were used for classification. The results indicated that the SVM model obtained the best recognition effect, in which the average accuracy reached 93.9 %. Furthermore, CNN was used to classify the adulterated milk powder. The results showed that the CNN model had better performance, and its average accuracy was 97.8%. In addition, the interaction process between CNN and spectral data was analyzed by visualization of CNN layers, which proved that CNN could effectively extract spectral features. Overall, those results demonstrated that LIBS combined with CNN was a simple, reliable, and high-accuracy method for the identification of milk powder adulteration.