Abstract: In response to the mandatory situation of passengers wearing masks on rail transit during COVID-19 pandemic period, a lightweight mask detection algorithm (Mask-Det) based on deep learning is proposed. First, the lightweight backbone network EfficientNet is used to extract image features. Then, a highly efficient feature fusion module is used to enhance the semantic information of the shallow feature map for detecting small targets. Finally, the algorithm is trained on the dataset of public scenarios, and then further optimized on the dataset of rail transit scenarios using transfer learning. Mask-Det algorithm has high detection accuracy, small model parameters, and fast detection speed. The algorithm can detect in real time whether passengers are wearing masks at various places, thus effectively alleviate personnel stress and improve passenger entry speed.