Abstract: The small target fault pixel ratio is low during aerial patrol inspection, which leads to the problem that their features cannot be effectively expressed and thus difficult to be accurately detected by traditional algorithms. Therefore, a transmission line small target fault detection network which integrates prior knowledge and attention model (named as PKAMNet) is constructed. Firstly, in order to solve the problem of low detection network accuracy caused by the lack of small target fault samples of transmission lines, the small target fault samples are prepared in the laboratory environment for data expansion, and the CoT module is integrated into the CSPDarkNet as the backbone network to extract the small target fault features; meanwhile, the weight and parameters of the feature extraction network are transferred to the backbone detection network to realize the prior knowledge transfer of fault features. Secondly, considering the low accuracy of the detection network caused by the weak saliency of the fault target image in the complex background, the efficient channel attention (ECA) model is embedded in the backbone network to improve the saliency of the fault target area, so as to improve the feature expression ability of the fault target in the complex background environment. Then, in order to solve the problem that the loss function degrades when the prediction frame coincides with the target frame, which makes the network difficult to converge, the network coordinate loss function is changed to Alpha-CIOU to improve the detection network accuracy. Finally, in order to verify the advantages of the proposed algorithm, it is compared with four classical algorithms on the data set collected by an inspection department. The experimental results show that the proposed algorithm has the highest detection accuracy, with an average detection accuracy of 94.2% and a detection speed of 48 Frame/s for images with a resolution of 1280×720.