There is a problem of workers who don't wear safety helmets as required as well as non-operating personnel entering a job site by mistake. Thus a deep learning-based safety helmet and voice recognition intelligent terminal algorithm is designed. First, for the detection of helmets, we use a human body key point detection model and a YOLO3 algorithm based on deep learning. The video file obtained by the smart camera is first used to extract the images of the on-site personnel using the human body key point model, and then the YOLO3 algorithm is applied to detect the situation of the on-site workers wearing helmets and send out warning messages for those who do not wear the helmet correctly. Finally, the practicality of the proposed model is verified through model training. This work is supported by the Science and Technology Project of China Southern Power Grid Company Limited (No. 037800k52190006).