Automated product and price information extraction from retail promotional flyers using YOLO and OCR

Abstract Promotional flyers are widely used by retailers to advertise products and prices; however, extracting structured and meaningful data from them remains a significant challenge. Their highly unstructured layouts, visual clutter, and diverse design styles make data extraction largely manual, time-consuming, and error-prone. This paper presents an automated approach for extracting products and prices from retail promotional flyers using deep learning–based computer vision techniques. The proposed pipeline integrates a YOLO-based object detection model for localising product and price regions with optical character recognition for extracting the associated text. The study focuses on visually complex Nordic retail flyers. Hundreds of flyers were curated and annotated from a SQL dataset provided by VG-LAB, capturing substantial variation in layout, typography, background noise, and image quality typical of Scandinavian retail advertising. To improve model robustness, a lightweight augmentation strategy was applied, including greyscale conversion and noise reduction. In addition, a scenario-specific fusion of YOLO-based object detection and OCR was designed and optimised for this context, enabling more reliable localisation and text extraction under challenging visual conditions. The approach was evaluated on a real-world dataset of weekly promotional flyers and showed reliable performance in detecting both product and price information. The system was also deployed through a FastAPI-based pipeline for real-time and scalable processing. This demonstrates its practical relevance for platforms such as VG-LAB’s Kupp application, where automated extraction can reduce manual effort and provide structured, searchable product information for consumers. In conclusion, the findings highlight strong potential for scalable and efficient retail flyer information extraction.