Face mask detection for covid-19 standard operating procedure by using deep learning

Abstract

In 2019, a new and highly infectious disease emerged in Wuhan, China, and quickly spread around the world. SARS-Cov 2 (also known as COVID-19) is the illness. The coronavirus COVID-19 pandemic is wreaking havoc on the world's health system, infecting over 180 million people, and killing over 3.8 million people. The virus transmission method was spread through respiratory droplets when an infected person coughs, sneezes or even speaks Even though a vaccine is available, there is still no effective treatment for this condition, and even if one is vaccinated, one can still be diagnosed. Therefore, the best approach to deal with it is to avoid it, and many medical experts have recommended wearing a face mask as one of the most effective ways to stop the virus from spreading. Aside from that, numerous countries throughout the world have enacted new laws or guidelines requiring individuals to wear face masks on a regular basis. However, some people continue to refuse to use a face mask when visiting public areas, especially in crowded places. As a result, stationing a security guard at the entry to monitor visitors appears to be the alternative. This approach, however, not only puts the guards in risk, but it also has the potential to cause overcrowding at the gate due to its inefficiency. Machine learning is undeniably the key to averting this downfall by minimizing direct human participation. Over the year, in field of image processing and computer vision, the spotlight was more focus on only face detection rather than face mask detection therefore the vulnerabilities of face mask detection technologies have not been properly addressed. Hence, the first objective of this paper is to implement the deep learning in image recognition for face mask detection. Next, the objective will be developing a system that able to detect whether a person is wearing a face mask or not by utilizing Convolutional Neural Network (CNN). In this project, the CNN architecture, MobileNetV2 is being utilised due to its low computational cost. A total of 3486 images of face masked and without face masked datasets are created from various online open-sourced datasets and fed to the model. Several optimizers, including SGD, RMSProp, and Adam, are evaluated to obtain the optimal network model. Finally, the Adam optimizer is chosen, and optimization techniques such as epoch size, batch size, and initial learning rate are gradually tuned and applied to the model. The validation accuracy could reach 99% throughout the tuning process, and the validation loss was decreased from 4.13 % to 2.86 %. The result model was then compared to another state-of-the-art CNN model, VGG-16, and the results reveal that the MobileNetV2 model did indeed utilise fewer computing resources, as it consumed 11% less memory, had a 5 times smaller result model, and took 5 times less time to train than that in VGG-16. When the model was put to the test with 50 real-life example images, the model able to achieve accuracy of 86% which the model able to detect the face in the image and correctly labelled it. In the end, the system able to detect face and distinguish the face with or without face mask and thus help in face mask detection to prevent the spread of COVID-19.