Wavision: Microwave-Sensing System for Food Safety
by Francesca Vipiana, Marco Ricci, Giovanna Turvani, Mario Casu, Jorge Alberto Tobon Vasquez by from the Wavision project of Politecnico di Torino
It probably happened to everyone to read news about food recall due to contamination issues, even causing a sense of loathing . The continuous growth in mechanized processes in food factories is one of the main causes of such occurrences. In addition, this contamination is undoubtedly a potential issue for a brand reputation, which could be heavily affected by this problem. But primarily, it must deal with consumers’ health: accidental ingestion of any foreign body might cause severe damage to the digestive system, particularly for children and seniors.
Indeed, employed inspection devices have some deficiencies, missing the detection of low-density materials, such as plastics, glass, or wood. The most powerful and complete systems employed in industries are based on X-Rays, whose detection principle grounds on materials densities. Moreover, this solution deals with ionizing radiation. It requires particular attention and lengthy procedures for authorization and careful training for operators to be accepted in a highly regulated industry, as food production is.
Wavision aims to solve the mentioned issue through an in-line system, adaptable to existent conveyor belts; it deals with low-power microwave signals, scattered by an antennas array surrounding the product to analyze. Its working principle is based on the dielectric features of the content: potential contamination will alter the microwave signals compared to a reference case, making it detectable by our system. The microwave radiations are completely safe for operators, and their low power level will not impact the investigated products. The antennas are low-cost printed circuit boards. Depending on the medium to inspect, their exact working frequency is designed as a trade-off between the maximum resolution and the best penetration. An analysis of the product’s dielectric characteristics, and in particular its loss, will set up this value. The optimal frequency will allow the signal to sufficiently penetrate the target, localizing potential intrusion all over its volume.
The antennas’ position is evaluated through an offline and preliminary analysis targeted to maximize the spatial coverage of the device. In addition, the acquisition speed is sufficiently fast to let the system collect real-time signals, inspecting each of the samples passing along the line.
Another offline procedure is thought for its development: the classification of samples is performed employing neural network implementation. It has to be fed adequately with signals collected during a training phase, in which both clean and contaminated samples get measured. Our studies demonstrated the robustness of such a classifier in the sense that it can generalize: it can recognize the presence of contaminants different from the ones measured in the training phase. This is a crucial aspect since the nature of potential intrusions cannot be precisely foreseen.
All this procedure can be performed once before installing the system; it is even flexible, in the sense that it is possible to potentially adjust it by quickly adding training samples to improve its robustness. Once adequately trained, the classifier can give an output of a current measured sample instantly. Our experiments showed a classification accuracy close to 100%.
Concluding, Wavision proposes a microwave-based inspection system to guarantee the expected safety from industrial products, ensuring no contamination. Furthermore, it is performed with a flexible design, easy to adapt to existing production lines, complementing the lack of currently adopted devices.
 — https://www.scientificamerican.com/article/the-8-weirdest-food-recalls-in-america/