Non-destructive Evaluation of Food Products by Microwave and mmW Imaging

Ensuring food safety and quality is a major concern in the food industry, as physical contamination can pose significant health risks to consumers, damage brand reputation, and lead to legal consequences. Common detection methods, such as X-ray inspection, have limitations, particularly in identifying contaminants like plastic, wood, and glass, which have low density and can be difficult to detect. This work explores an alternative solution based on microwave sensing, introducing a novel approach to contamination detection. The proposed system leverages low-power, non-ionizing microwave signals to identify foreign bodies without compromising food integrity or requiring extensive modifications to existing production lines. It offers a cost-effective and real-time inspection method, capable of operating in-line without interrupting the manufacturing process. The detection principle relies on analyzing how microwave signals interact with different materials, taking advantage of the dielectric contrast between contaminants and food products. A set of antennas surrounding the target captures signal variations, which are then processed to determine the presence of foreign objects. The system is designed to acquire data efficiently while maintaining compatibility with the speed and constraints of industrial food processing environments. Detecting contaminants in food and beverage products using scattering parameters involves solving an inverse problem, which is nonlinear and ill-posed. This process is computationally expensive and may not be suitable for real-time, in-line detection. In this research, we integrate Machine Learning (ML) techniques to overcome the challenges of the inverse problem and to automate the classification process. We investigate the robustness and effectiveness of different classifiers, such as Support Vector Machines (SVM) and Multi-Layer Perceptron (MLP) networks, by training them on large-scale datasets collected from experimental trials. These models learn to distinguish between uncontaminated and contaminated food items. The classifiers successfully identifies a variety of foreign materials, including different types of plastics, glass, and wood, demonstrating high accuracy across thousands of test cases. Furthermore, we extend our research to millimeter-Wave (mmW) imaging, investigating the integration of mmW systems with ML tools for nut inspection. The promising results achieved using mmW imaging and ML classification in agrifood applications, particularly for soft fruits like apples and peaches, inspire us to evaluate the effectiveness of this approach for more challenging cases—specifically, in-shell seeds such as almonds and walnuts. The results obtained in this thesis highlight the potential of the microwave/mmW-based system as a robust, scalable, and efficient solution for real-time food contamination detection and agrifood inspection. By integrating microwave sensing with machine learning, this approach offers a powerful alternative to traditional inspection methods, improving food safety and quality in industrial settings