(PTFS05-02-25) Using Artificial Intelligence and Computer Vision to Detect and Monitor Unhealthy Child-Directed Food and Beverage Marketing

Objectives: Childhood obesity and diet-related diseases are urgent global public health challenges, driven in part by children’s continued exposure to persuasive marketing of unhealthy foods. Manual evaluation of child-targeted food marketing is resource-intensive. This study leverage AI to identify child-targeted food products, classify marketing elements on food packaging, and explores the practical application of AI within a specific food category to assess the potential impact of proposed Canadian marketing-to-children regulations.

Methods: Image classification algorithms were trained on 8,283 manually labeled food package images, validated using a child-appealing packaging coding tool. Three machine learning algorithms (i.e., kNN, SVM, and CNN) were used to classify food package images targeted at children. YOLOv10 object detection were used and fine-tuned to identify specific child-targeted marketing features. Model performance was evaluated using accuracy, F1 score, AUC, and confusion matrix. The AI approach was applied to breakfast cereals to assess compliance with proposed Canadian regulations and relationship with nutrition quality and price.

Results: A total of 22 child-directed marketing techniques featured were identified and annotated. The CNN model outperformed kNN and SVM, achieved 0.92 accuracy and 0.98 AUC in identifying child-targeted marketing on food packaging. YOLOv10 object detection model showed varied performance across child-directed marketing features. Among breakfast cereals, 35.1% were identified to display child-directed marketing features that would be restricted under Canadian proposed marketing-to-children regulations. Products with child-directed marketing features tended to be less expensive, though not significant (p=0.072).

Conclusions: This study introduces an AI-driven strategy for identifying and categorizing child-directed marketing on food packaging, with potential extensions to social media and online video content for broader applications. These scalable methods support efficient monitoring of compliance with marketing-to-children policies and provide evidence-based insights to reduce children’s exposure to unhealthy food marketing to improve health outcomes.

Funding Sources: This work was funded by the Data Sciences Institute Catalyst Grants and the CIHR project grants.