COVID-19 has created a global serious health hazard with far-reaching consequences for society, our perceptions of the world, and how we live our daily lives. As a result, the World Health Organization recommended the use of face masks and social isolation to help reduce the rising number of infections. However, subsequent research has revealed that face masks alone can be ineffective, particularly in crowded settings or hospitals. Face shields can also be used in addition or as an alternative for face masks because they are indefinitely reusable and can be washed with soap and water or standard disinfectants. Because most detectors for fighting COVID-19 only focus on the face mask alone, we proposed a transfer learning model by fine-tuning the pre-trained MobilenetV2 architecture, to detect, recognize, and distinguish faces with shield, mask, and those without either. This study applied a standard image recognition pipeline, which is comparable to that used by most traditional recognition programs. In doing this, we first downloaded and scrapped images from search engines to form our dataset, we then pre-processed the images by the application of image augmentation to address the limited availability of the dataset for a better training and validation. After which a multi-class detection system was accomplished. The results of the study achieved 98 percent accuracy on the validated dataset. It is therefore recommended that this model can be improved to capture all forms of face covering and be integrated into CCTV cameras for its detection in important places like hospitals.