Developing innovative food safety solutions: an open-source toolkit for characterising foods

The food industry constantly grapples with the challenge of ensuring the quality and safety of fresh or transformed foods (e.g., different types of meat), which naturally vary in their characteristics and perishability. At present, factories continue to monitor food quality using destructive chemical analyses that are also slow and often highly technical. For example, for fish products, technicians evaluate product firmness by touch and carry out chemical analyses that will yield an overall average the following day at best. Furthermore, these analyses require ground samples.

What are the advantages of hyperspectral imaging?

Hyperspectral imaging (HSI) can detect "invisible” food characteristics (e.g., water content, fat content, trait changes) using spectral bands of visible and infrared light. Rapid and non-destructive, it is an alternative technique for monitoring food quality over the course of production. However, researchers and industry stakeholders have yet to widely adopt HSI because it generates highly complex data and proprietary software for analysing HSI data is costly.

INRAE joined forces with the University of Tokyo and headed up the development of a transparent, open-source tool with the aim of democratising and standardising HSI use.

To begin with, HSI can be employed to track chemical changes in foods (e.g., oxidation, colour variation): different types of spectral cameras detect visible and infrared wavelengths, yielding pixel-by-pixel information that is used to map a product’s surface. The researchers applied HSI to ripening sausages and were able to spatially quantify sausage drying and oxidation over time. The technique clearly revealed water loss dynamics, which started in the periphery of the sausages. This information could be used to optimise the sausage ripening process.

Predicting compound presence

The toolkit’s main innovation is that it can produce a map that predicts the presence of a target compound via an efficient form of AI (i.e., one that uses less energy and fewer computing resources). For example, the toolkit can predict the presence of salt, a compound that is "invisible” to the spectral cameras. Specifically, when used on trout fillets, the toolkit’s algorithm reliably predicted the presence of salt 98% of the time, by correctly identifying the indirect effects of salt on surface water dynamics. These precise spatial predictions could tremendously boost quality control. While traditional analyses provide an estimate of average salt levels, this toolkit reveals small-scale heterogeneities. Identifying conditions that could promote microbial contamination is essential to ensuring food safety. For example, bacteria could develop in highly localised zones with too little salt.

The toolkit can be used with all types of foods and biological tissues. A basic version is currently available for use as is, but readymade software might be developed over the course of a future project. The long-term objective is to generate a tool that can be deployed during industrial food transformation or preservation processes. For example, one application could be to provide an early warning system for bacterial growth in food.