Development of an AI-based Application for Counterfeit Medicine Detection in the Nigerian Drug Market

Counterfeit medicines pose a significant threat to public health worldwide, creating a necessity for detection systems to ensure consumer safety. This research focuses on developing a web-based application using computer vision and Natural Language Processing (NLP) techniques for counterfeit medicine detection. The application integrates logo detection, Optical Character Recognition (OCR), and spell-checking functionalities to validate the authenticity of pharmaceutical products and packaging. By utilizing transfer learning on the YOLO-NAS model and leveraging the Microsoft Common Objects in Context (COCO) dataset, a custom logo detection model was trained to identify approved brands. The OCR functionality utilizes the Google Vision API for accurate text extraction, followed by a Named Entity Recognition model to filter out non-English words and names of places before spell-checking. The custom logo detection model was trained for 200 epochs, achieving an overall mAP of 79.3%, Precision of 85.3%, and Recall of 75.3%. This indicates that the model when integrated into the application is optimized for detecting counterfeit medicine efficiently. The application features a user-friendly interface with three main pages: the home page, the authenticated successfully page, and the failed to authenticate page, providing intuitive navigation and feedback based on authentication results. Equally, the Series of evaluations by the 16 undergraduate students suggests that the application can be reliable and effective in real-world scenarios.