Abstract: Due to the increase in the scale of the power grid, the large number of applications of helicopters and unmanned aerial vehicles, the number of aerial images produced has increased dramatically. Among them, due to the large number and small size of bolt defects, accidents caused by bolt defects on transmission lines occur frequently. In addition, the existing bolt defect classification methods for transmission lines are limited to surface feature extraction while ignoring problems such as correlation between targets and great influence from complex environments. In view of the above, this paper proposes to use the association between bolts and nuts to form a bolt and nut pair, and then use the convolutional neural network to extract the bolt and nut pair feature initialization graph network nodes and combine the prior knowledge of the bolt and nut pair to indicate the bolt and nut pair defects and bolt and nut pair. Associate the semantic objects, and use this to establish a bolt and nut pair knowledge map to guide bolt and nut pair defect classification. If the lack of bolts and bolt-related defects on power transmission lines are divided into bolt and nut pair defects, a coarse-level defect data set and a fine-level defect data set are established. Through the use of bolt and nut pair knowledge graph to guide the defect classification experiment of bolt and nut pair, and to verify the effectiveness and feasibility of bolt and nut pair knowledge graph. The experimental results show that the bolt and nut pair knowledge graph realizes the effective use of the bolt and nut pair prior knowledge, and completes the efficient classification of the bolt and nut pair coarse and fine defects.