System for the automated monitoring of sugar, fat, and salt in processed foods
Search for relief in data collection with artificial intelligence (AI)
Within the context of the national reduction and innovation strategy, the MRI conducts an annual product monitoring to survey the contents of energy, sugar, fat, and salt in processed foods over time. The data is collected manually in a very laborious and time-consuming process.
The BMEL-funded cooperation project RePro, completed in July 2024, sought to develop an efficient system for gathering and comparing online data on the energy and nutrient content of processed foods over time. The aim was to emulate the manual product monitoring conducted by the MRI in a maximally automated process.
Intended functions
The components of the RePro system:
- RePro crawler (data search)
- RePro wrapper (data extraction)
- RePro storage (data storage)
- RePro analysis (data analysis)
- RePro dashboard (data management)
With the help of an internet meta-search engine (RePro crawler), product data are collected from predefined online data sources like websites of food manufacturers and supermarkets. Relevant data such as product name, manufacturer, ingredient list, and energy and nutrient content of the processed foods can be extracted (RePro wrapper) and stored (RePro storage).
Within the data analysis (RePro analysis), foods are categorised into predefined product groups, for example, and energy and nutrient contents are compared over time. The analytical results can be organised on an internal platform for data management (RePro dashboard). The two cooperation partners, the software company snoopmedia GmbH (project coordinator) and the AI company elevait GmbH & Co KG, were responsible for the technical implementation.
Research contribution of the MRI
The Institute of Nutritional Behaviour at the MRI provided nutrition and food science expertise. Specifically, this meant defining the monitoring requirements of the RePro system and evaluating the results of the automated data collection against those of the manual product monitoring.
Results of the evaluation
To evaluate the RePro system, the results of the automated data collection were compared to those of the MRI manual product monitoring of 2022. The RePro system recorded approximately half as many products as the manual product monitoring. The technical implementation of capturing product information from the supermarket websites proved to be easier than from the websites of the food manufacturers. Due to the heterogeneity of the products, the automatically generated data set had to be amended manually.
Overall, with the RePro system the data collection and processing effort was lower compared to the manual product monitoring. The automated process, however, covered fewer sources. In the RePro system the products were primarily categorised using a rule-based approach. The tested AI technologies for image recognition need to be developed further to become functional. The distribution of energy and nutrient content data was mostly similar in the automatic and the manual product monitoring.
Future approaches with new technologies
The RePro project has generated useful insights into where and how at least parts of the product monitoring can be automated. In the meantime, important technological innovations have taken place that should be checked for their usability. For example, research efforts should go into whether the use of new and improved AI applications such as Large Language Models can further advance the automation process.
Other monitoring projects:
- Product monitoring
- Monitoring of the reformulation of processed food (Best-ReMaP )