Real-Time Pill Detection and Recognition Framework Based on a Deep Learning Algorithm

Abstract

One of the most common pharmaceutical dosage forms is oral pills, including tablets and capsules. Oral drugs are fairly stable and simple to deliver compared to other dose forms, such as syrups and injections. However, misidentifying medicines is not uncommon, whether inside healthcare institutions or after the pills have been delivered to patients. The authors propose a new framework for real-time pill detection and recognition using deep learning algorithms. The proposed framework comprises three main models: firstly, pill strip identification using the YOLOv5; secondly, the text detection model locates the text information in the pill strip image; and finally, the text recognition module recognizes the localized text. The recognition module greatly assists the proposed framework by identifying the text information such as pill name, expiring date, price, etc. This work's training and testing images are captured from various perspectives and varying illumination levels. Despite suboptimal image quality, the proposed framework obtained good recognition accuracy.