Sensor Approaches for the Non-Destructive Assessment of Food Safety and Authenticity

Abstract

Ensuring access to high-quality, safe food is typically the responsibility of national governing bodies overseeing food production and distribution, and is usually secured by enforcing compliance with national food standards at the producer level. However, recent and historical food safety transgressions have illustrated that conventional food standards frameworks have fallen short in some respects: conventional batch-based compliance testing has been shown to result in delays between identification of problems and corrective action, resulting in product recalls, batch destruction, and reduced consumer trust. In severe cases, delayed responses have exposed consumers to unsafe products before action was taken, leading to harm. Modern, high throughput sensor technologies, coupled with appropriate informatics support have the potential to improve the responsiveness of current food standards frameworks, offering a means to monitor the food production and distribution chain in higher resolution, while allowing for close-to-real-time detection of emergent issues, before exposure to the public and before incurring costs for the food business operator. This work shows a two-part appraisal of this approach. First, mathematical models were developed to assess the accuracy and reliability of candidate sensors for prediction of food safety and quality parameters in meat products, both in isolation and in a sensor fusion context. Second, an online platform was constructed to support development and deployment of predictive models built using multiple channels of high-throughput sensor data in a manner enabling performance evaluation and real-time decision support capabilities. The methods and platform developed in this work demonstrate that employment of high-throughput, minimally invasive sensor technologies may supplement existing monitoring practices when coupled with suitable machine learning pipelines, and such pipelines, including those integrating multiple streams of multi-modal sensor data, can be deployed efficiently in a decision support capacity.