The PortionSize Ed Mobile App Demonstrated Feasibility, Acceptability, and Preliminary Efficacy for Improving Diet Quality Among Early Adolescents in SNAP-Ed Hawaii

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 (p0.072).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-tochildren policies and provide evidence-based insights to reduce children's exposure to unhealthy food marketing to improve health outcomes.