Automatic Pill Identification System Based on Deep Learning and Image Preprocessing

The pharmaceutical industry annually distributes thousands of different prescription medications. However, the high volume of available medication increases the likelihood of errors during the dispensation process. These errors can result from distractions, incorrectly filed prescriptions, or similarities between pills. This paper proposes an automated system that employs computer vision to identify pills, aiming to enhance pill identification for pharmaceutical workers and consumers and reduce dispensing errors. The proposed approach involves creating a deep learning model using Keras, preprocessing pill images through an image data generator, and leveraging tools such as OpenCV and Paddle OCR to identify critical aspects of a pill, including its shape, color, and imprint. The ultimate goal is to develop a real-time pill identification system that can seamlessly integrate with video cameras, thereby facilitating smoother operations in high-volume medication dispensaries.