Safety in the construction industry has become a hot issue for a long time, but it is always difficult to eliminate hidden dangers in construction safety. With the development of drones and artificial intelligence technology, the application of image acquisition and identification technology to the field of building construction safety prevention and control is one of the current research hotspots. High-altitude fall accidents are the type of accidents with the largest proportion of construction safety accidents. The lack of safety guardrail is part of the main reasons for frequent high-altitude fall accidents. This paper uses a deep learning framework to build 5 mainstream convolutional neural network models to train and test 3600 construction site image data sets collected by UAV(unmanned aerial vehicle). The experimental results demonstrate that the accuracy of the safety guardrails in the five models for identifying security detection has reached more than 90%. After comparative analysis, it is concluded that the Mobilenet model has the best recognition effect for the border guardrails. The research results verify the effectiveness and feasibility of the image recognition CNN algorithm and UAV technology for quickly identifying potential safety hazards at the construction site, which will help eliminate potential safety hazards at the construction site and improve the safety management level of the construction industry.