Development of a Deep-Learning Pipeline to Detect and Locate Contaminants of Industrial Products via non-Invasive Microwave Signals

Nowadays, food companies produce large volumes of packaged food products to satisfy the demand of a population that keeps increasing in number. One of the biggest challenges these enterprises must face is how to scale accurate contaminant detection methods. In this paper, we improve atop existing microwave-sensing techniques for food contaminant detection by adding Machine Learning (ML) techniques. We consider a greater variety of common contaminants in the food industry with respect to the literature. Moreover, we enhance the data collection phase and propose a Graph Neural Network (GNN)-base approach to detect the position of the contaminant. We show that this enhanced data combined with the proposed ensemble of ML algorithms outperforms the accuracy of the detection with respect to the state-of-the-art approaches.