AUTOMATIC DRUGS RECOGNITION MODEL USING CONVOLUTIONAL NEURAL NETWORK

Abstract

Issuing the right place or the number of the different drugs in one place can be a controversial problem in many pharmacies. In the health care system, medication errors can be one of the biggest and most important issues that affect safety. One of the reasons this kind of error happens is the “human factor” whose reason can be insufficient knowledge or fatigue. The use of a robotic system to manage drugs in the hospital is highly recommended during the COVID-19 pandemic. To make a robotic system manage drugs in the hospital. First, it must recognize the drugs in the hospital. To do that, this paper proposed drug recognition using deep learning. Developing a model that can recognize the drug as well as its location can help to eliminate almost all the errors. This study aims to understand between different materials which one is a drug, how many drugs are there, and where the location is. We employed a convolutional neural network with VGG architectures as well as different activation functions to develop a model with the highest accuracy. All images were categorized into 3 different classes; Bottle packaging, Blister packaging, and Syringe. To create a dataset, the images were converted using data augmentation. As VGG is very popular for this kind of problem, we expect to develop a model with a VGG structure with an accuracy of more than 85%. Moreover, LeRu and Sigmoid seem to have better performance as activation functions compared to others. Identification and localization of different medicine and drug with our proposed model is with high accuracy and precision. This model can be implemented in a mobile app to scan the images from the different objects and recognize drugs and medicines to help the workers of the healthcare system to accelerate related affairs and reduce human error.