YOLO-ARGhost: a lightweight face mask detection model

• Published in: The Journal of supercomputing Vol. 80; no. 3; pp. 3162 - 3182
• Main Authors: Xu, Ruihang, Wang, Peisen, Li, Xuanjing, Nie, Ruifeng
• Format: Journal Article
• Language: English
• Published: New York Springer US 01.02.2024; Springer Nature B.V
• Subjects: Compilers; Computer Science; Face recognition; Feature extraction; Industrial development; Interpreters; Lightweight; Masks; Modules; Performance enhancement; Performance evaluation; Processor Architectures; Programming Languages; Respiration; YOLO-ARGhost; Face mask detection; Lightweight model; YOLOv4
• ISSN: 0920-8542; 1573-0484
• DOI: 10.1007/s11227-023-05588-3

Industrial development can bring huge economic benefits to the country and society. However, it has also caused serious environmental pollution, leading to serious health problems and medical burdens for people, and is often accompanied by the emission of polluting gases. Many companies specify that masks must be worn at work to prevent the inhalation of harmful gases. Quickly detecting whether workers are wearing masks has emerged as a topic of some importance. However, existing face detection networks have limited detection accuracy and considerably reduced performance for masked faces. Hence, most state-of-the-art face mask detection technologies are based on deep learning. In this study, we developed a new feature extraction module called Attention Residual Ghost Module based on attention mechanisms and residual structure. To improve performance, we construct CSP-ARG and ARG-PANert. We then fused attention between the two to obtain a new type of lightweight face mask detection model. The results of an experimental evaluation of the performance of our proposed approach on the public AIZOO and FMDD datasets showed that it achieved accuracy values of 93.4% and 89.3%, respectively, in terms of mAP.