At present, the situation of epidemic prevention and control is grim. Real time and rapid mask wearing detection in crowded places can effectively reduce the risk of virus transmission. Aiming at the low efficiency of manual detection, a lightweight mask wearing detection algorithm based on YOLOv3 is proposed. ShuffleNetv2 is used to replace the original backbone feature extraction network to reduce the amount of network parameters and computing power consumption. SKNet attention mechanism is introduced into the feature fusion network to enhance the ability of feature extraction at different scales; CIoU is used as the boundary box regression loss function to further improve the detection accuracy. Experiments on the constructed face mask detection data set show that, compared with the original YOLOv3, the proposed algorithm improves the detection speed by 34FPS while maintaining high detection accuracy, and effectively realizes accurate and fast mask wearing detection. Compared with other mainstream target detection algorithms, the algorithm also has better detection effect.