Due to the need for epidemic prevention and control and the performance superiority of the yolov3-spp network, this paper mainly uses yolov3-spp, which is a Yolo-based pruning network to detect epidemic prevention and control behaviors. Specifically, firstly, the epidemic prevention and control dataset suitable for the Yolo algorithm is constructed. Then, on the premise of saving cost and easy deployment, the model compression experiment is carried out. Through comparative experiments, we found that: (1) Yolov3-spp can be applied to the epidemic prevention and control dataset and detect the epidemic prevention and control behavior. (2) The Recall and mAP of the model before and after pruning are close, but the inference speed, model size, and model parameters after pruning are about 2 times, 10.7% and 10.8% respectively of those before pruning. (3) Balancing the requirements of model performance, inference speed, and computing power, our method is more suitable for deployment on edge devices than yolov3-spp, yolov4-tiny, and yolov4. These experimental results show the feasibility, effectiveness, and adaptability of the pruning method described in this paper.