Detection and Classification of Urea Adulteration in Milk with Deep Neural Networks

Abstract

Milk is a major food constituent. However, the existing discrepancy between milk demand and supply leads to adulteration, which can be dangerous since it causes detrimental effects on health implicating lethal diseases. Although classical methods for adulteration detection are very accurate, their implementation requires skilled technicians as well as expensive and sophisticated instruments. These reasons trigger the need for improved techniques in uncovering adulteration. Urea is a natural component in milk and accounts for a substantial share of adulteration in the non-protein content of milk. The current research proposes and employs a sensor system utilizing the Electrical Impedance Spectroscopy (EIS) method to determine the presence of urea. The classification system was developed using different machine learning algorithms. Three classifiers, Extreme Gradient Boosting (XGBoost), Extreme Learning Machines (ELM), and Deep Neural Networks (DNN) were considered for various levels of urea adulteration. Milk samples were assessed by deploying the developed EIS sensor assembly and the results derived were employed in the training of the machine learning algorithms. The estimated classifiers displayed promising outcomes, involving up to 98.33% classification accuracies, outshining frequently used existing learning approaches like logistic regression.