A Deep Transfer Learning and Fine-tuning Model for Face Mask Detection and Classification

The face mask detection model aimed to minimize the physical interaction from person to person to prevent transmission of the virus. The gathered legally licensed face mask images for training and testing were categorized as Bare, Cloth, KN95, and Surgical. Since the collected images used for dataset were limited and had variable sizes since they come from different sources, Keras from Tensorflow is extremely beneficial when it comes to pre-processing and data augmenting images. The fine-tuning technique as a popular deep neural network transfer learning strategy was applied few rounds of training to the parameters of a pre-trained model to adjust it to enhance its accuracy and performance. As a consequence, among the three convolutional neural networks applied to each model, the MobileNetV2 architecture demonstrated very high performance specifically in accuracy, recall and f1 score even if it is not fine-tuned and data augmented.