A Deep Learning Approach for Detecting Face Mask Using an Improved Yolo-V2 With Squeezenet

• Published in: 2022 IEEE 6th Conference on Information and Communication Technology (CICT) pp. 1 - 5
• Main Authors: Kwaghe, O. P., Gital, Abdusalam Yau, Madaki, A.G, Abdulrahman, Mustapha Lawal, Yakubu, Ismail Zahraddeen, Shima, Iosun Stephen
• Format: Conference Proceeding
• Language: English
• Published: IEEE 18.11.2022
• Subjects: Convolutional Neural Network (CNN); Region based Convolutional Neural Network (RCNN); Region Proposal Networks (RPN); Single Shot Detector (SSD); SqueezeNet; You Only Look Once (YOLO)
• DOI: 10.1109/CICT56698.2022.9997956

Face mask detection has become a critical issue in security and Covid-19 prevention. In this regard, the YOLO V2 network has demonstrated outstanding performance. The YOLO V2 on the other hand, employed Darknet as a feature extractor. However, as compared to Darknet, SqueezeNet allows us to reduce model size while reaching or surpassing the highest accuracy score. SqueezeNet is designed to have lower parameters that can be more readily stored in computer memory and transferred across a computer network. As a result, in this study, we recommended enhancing the YOLO network by replacing Darknet with Squeezenet. Compared to other existing face mask recognition systems that use the standard YOLO V2 algorithm, this improves overall performance in terms of model size and accuracy. As a result, this study proposed a rapid face mask detection model by improving the existing YOLO V2 network architecture by employing logistic classifiers and SqueezeNet for multi-label classification using FMD and MMD face-masked dataset. The model was evaluated on MATLAB 2021 against state-of-the-art approaches. The proposed model outperforms previous algorithms by attaining a good accuracy value of 81% and a recall value of 99.99%.