Comparative Analysis of Deep Learning Techniques for Facemask Detection

ABSTRACT

Pandemic caused owing to widespread of corona-virus has changed our lives upside down. Covering the face area particularly nose and mouth is the prime need of the hour. Any negligence of not wearing the mask or incorrectly wearing the mask can be hazardous. This necessitates the need of understanding the real importance of wearing the mask appropriately in order to avoid the spread of Covid 19. Knowing the present population of the country, manual monitoring of the individuals is quite difficult. So, this research puts forward the use of deep learning techniques for automatic facemask detection using techniques such as capsule network, ResNet50, Mobile-Net architecture, and Convolution Neural Network. The techniques are validated on the merged dataset taken from MaskedFace-Net dataset and Kaggle (publicly available) based on the performance measures namely accuracy, precision, recall and F1-score. Amongst all, the results showed that capsule neural network achieved superlative performance with the accuracy of around 99% in comparison to other aforesaid deep learning techniques. © 2022, Springer Nature Switzerland AG.