Abstract: The fault detection of the bolts on the transmission lines is of great significance to the safe and reliable operation of the power system. The bolts in the inspection images look unobvious in feature and small in size, which brings challenges to the research of their fault detection. With the development of helicopter and unmanned aerial vehicle inspection technology and the edge calculation, the traditional inspection image processing methods can no longer meet the needs of the real-time inspection. In this paper, a bolt fault detection system based on deep learning is proposed. Using the principle of hierarchical detection, the connection parts of the bolt fault are detected with single shot mutibox detector (SSD) firstly. Secondly, by cutting the other parts the proportion of bolts in the detection image increases and the data set is expanded with the data enhancement method. Finally, the Yolov3 is used to detect the bolt faults. The edge computing devices are mounted on helicopters or unmanned aerial vehicles to realize real-time detection of the bolt faults. The inspection images under different light intensities are detected to verify the robustness of the system. The experimental results show that this method can effectively and accurately detect the bolt faults in the inspection images.