Image Classification Using Deep Learning Architecture – EfficientNET for COVID-19 Medicine Consumption

Abstract

The Coronavirus disease 2019 (COVID-19) pandemic has impacted the world like no other pandemic in history. COVID-19 has progressed rapidly affecting health care in the global community. In this paper, we study the Image classification approach for confirming the consumption of medicine by people. The motivation for this research is to reduce contact with the people, without having them in physical observation and help minimize the spread of COVID-19. We used EfficientNet, a Deep Learning Convolution Neural Network (CNN) [7] architecture for systematical use of model scaling and balancing network depth, width, and resolution for better performance of our model. In this paper, we used the data augmentation technique to avoid overfitting. We used image hashing to compare and remove relatively similar images as data cleaning step and demonstrate the effectiveness of this approach by observing the classification accuracy. We used cloud computation to train the model. This trained model will be helpful to achieve the goal. In this study, we developed a model that could successfully detect the image and confirm the pill consumption by patient with high accuracy rate. This study will help medical professionals, especially during COVID-19 pandemic, in remotely monitoring the patient’s dosage and determine appropriate action in case of non-compliance.