A novel machine learning scheme for face mask detection using pretrained convolutional neural network

Mater Today Proc. 2022:58:150-156. doi: 10.1016/j.matpr.2022.01.165. Epub 2022 Jan 21.

Abstract

Corona virus 2019 (COVID-19) erupted toward the end of 2019, and it has continued to be a source of concern for a large number of people and organizations well into 2020. Wearing a face cover has been shown in studies to reduce the risk of viral transmission while also providing a sense of security. Be that as it may, it isn't attainable to physically follow the execution of this strategy. This proposed system is built by pretrained deep learning model, Vgg16. The proposed scheme is easy to implement and use all the layers in vgg16 model and train only the last layer called fully connected layer, which reduce the training time and effort. The proposed scheme is trained and evaluated using two Face mask datasets, one having 1484 pictures and the other with 7200. For a smaller dataset, augmented pictures were utilized to enhance accuracy. The suggested model is tested on unknown pictures, and it correctly predicts whether the image is wearing a mask or not. The proposed scheme gives accuracy 96.50% during testing in small dataset. The model gives accuracy in medium dataset is 91% during testing. By using vgg16 pretrained model and image augmentation in the dataset improves performance and gives a high accuracy.

Keywords: Face mask detection; Image augmentation and machine learning scheme; Transfer learning; Vgg16.