Face Mask Detection using an Automated Hybrid Deep Learning Method in the COVID Scenario

ABSTRACT

Currently, in light of the health catastrophe due to the COVID-19 which has been unfolding all over the world. Wearing a defensive mask has ended up a substitute normal. Face recognition technology is most commonly implemented for surveillance and other applications. Traditional machine learning classifiers as well as deep transfer learning classifiers have been used to accomplish the face mask detection mechanism. In this paper, two hybrid deep learning models MobileNetV2-SVM and MobilNetV2-KNN has been proposed for the task of face mask detection. The models involve two processes feature extraction and classification. For initialization, the MobileNetV2 pre-trained weights from ImageNet were employed, and during training, data augmentation and resampling were applied. By integrating the model with an SVM classifier and a KNN classifier, the model is further refined, creating hybrid models that are effective in terms of processing. The Kaggle dataset of 45000 images (22582 images are masked and 23423 images that are unmasked) of the proposed model/system is trained using MobilenetV2 and classified using SVM and K-NN algorithm in different models. Various machine learning frameworks were used like pandas, TensorFlow, Keras and NumPy. The accuracy achieved by the SVM model is 98.17% and 95.22% accuracy are achieved by using the K-NN classifier. © 2022 IEEE.