In the context of the policy debate and initiatives to address misleading practices that suggest to consumers that products marketed under the same brand and in the same or similar packaging have the same composition or characteristics across different Member States when this is not the case (often referred to as ‘dual quality’ (DQ)), the European Parliament has emphasised the importance of establishing a permanent monitoring system to track products sold on the single market. This feasibility study aims to explore the effectiveness of a crowdsourcing mobile app (Food-Checker) as a tool to monitor the occurrence of DQ. The app uses artificial intelligence (AI) technology and was tested using an awareness campaign and incentives to motivate users to submit product images. The app was piloted in five Member States (Germany, France, Italy, Poland and Romania) between September 2022 and March 2023. The study results demonstrate that Food-Checker can effectively utilise AI technology for monitoring DQ, particularly for capturing and extracting information from product packaging. Although there was a substantial awareness campaign, users’ engagement with the app was limited. In this regard, monetary rewards were found to be more effective than behavioural interventions. As a result of low user engagement and a small sample size, AI could not be fully trained in certain functions relevant to identifying DQ − such as identifying the same or similar branded products, distinguishing different product versions within a Member State and detecting DQ practices. The app could be a viable solution for monitoring DQ if the AI behind the app is further trained to enhance its performance and effective engagement tools are developed.
Authors
Joint Research Centre, European Commission, Di Marcantonio, Federica, Nedelcu, Bogdan Radu, Padiu, Bogdan, Rebedea, Traian, Barreiro-Hurle, Jesus, Ciaian, Pavel
- Catalogue number
- KJ-NA-31-855-EN-N
- Citation
- European Commission, Joint Research Centre, Di Marcantonio, F., Nedelcu, B., Padiu, B. et al., Food-checker â A mobile-based crowdsourcing application for âdual qualityâ of food , Publications Office of the European Union, 2024, https://data.europa.eu/doi/10.2760/242232
- DOI
- https://data.europa.eu/doi/10.2760/242232
- ISBN
- 978-92-68-12870-1
- ISSN
- 1831-9424
- Pages
- 88
- Published in
- Belgium
- Themes
- Information technology and telecommunications , Food technology and food safety
Table of Contents
- Abstract 5
- Executive summary 7
- 1. Introduction 15
- 2. Methodology 17
- 2.1. Artificial intelligence for image (front-of-pack design and label) recognition 17
- 2.1.1. Image classification 18
- 2.1.2. Instance segmentation 20
- 2.2. Product composition data extraction, translation and comparison 23
- 2.2.1. Instantaneous text translation 23
- 2.2.2. Information extraction process 23
- 2.2.3. Matching and comparison process 28
- 2.3. Mobile app interface and alert system 35
- 2.3.1. News 35
- 2.3.2. Proposals 36
- 2.3.3. Missions 37
- 2.3.4. Scan button and settings 37
- 2.3.5. Settings 37
- 2.3.6. Illustration of image-scanning steps 37
- 2.3.7. Information storing system 39
- 2.3.8. Results reporting 40
- 2.3.9. Alert system 41
- 2.3.10. Registration and data protection 45
- 2.4. Dedicated web page 45
- 2.5. Dissemination and awareness campaigns 47
- 2.5.1. Awareness and download campaign 49
- 2.6. Incentive mechanisms and rewards 54
- 2.6.1. Non-monetary extrinsic motivation: badges and points 56
- 2.6.2. Monetary incentives 58
- 3. Results 59
- 3.1. Performance of the artificial intelligence behind Food-Checker 59
- 3.1.1. Performance of artificial intelligence in object detection and image recognition 59
- 3.1.2. Performance of artificial intelligence in product composition comparison 62
- 3.2. Performance of the advertising campaigns and incentive mechanisms 63
- 3.3. Alert system 68
- 4. Operational costs of the Food-Checker app 71
- 5. Conclusions 73
- References 77
- Abbreviations 80
- List of boxes 81
- List of figures 82
- List of tables 83
- Annexes 84
- Annex 1. List of sampled products (number of samples) 84
