With the spread of the novel coronavirus in the world, COVID-19 is causing a global health crisis with countries already aware of the importance of the epidemic. According to the World Health Organization (WHO), wearing a mask is a necessary strategy for the prevention of COVID-19 infection. Due to this guideline, governments around the world are constantly asking people to wear masks in public places. However, it is currently inefficient to manually monitor if people are wearing a mask in public. Thus, to resolve this issue, in this paper we proposed a new robust and accurate model called COMASNet modified from the object recognition method (YOLOv5) to automatically detect if people are wearing a mask. Extensive experimental results on the proposed model gives us an accuracy of 98.1%. Besides this, our method has great advantages in identifying masks in a crowd compared with other models.