In view of the small number of samples in the current wearing mask data set and the limited hardware conditions, this paper proposes a lightweight YOLOv2 mask wearing detection method based on transfer learning. Based on the YOLOv2 target detection method, this method uses the MobileNetV2 of parameter transfer learning as the feature extraction network, which simplifies the network model and improves the training speed. The pre trained MobileNetV2 feature extraction network and YOLOv2 target detection network are combined to form a mask wearing detection network model. This paper collects and establishes a data set of 1000 pictures of face wearing masks to train and test the network model. The experimental results show that compared with YOLOv2 and SSD300 models, the average accuracy of mask wearing detection of MobileNetV2-YOLOv2 model is improved by 3.8%, 2.7%, and the detection speed increased by 2.5 times and 2.4 times. Moreover, under the condition of insufficient light and dense detection, MobileNetV2-YOLOv2 can still effectively detect mask wearing, which has better detection effect and stronger robustness than R-CNN and Faster-RCNN.