In the process of power system inspection, manual inspection is difficult and poses safety hazards. The unmanned aerial vehicle platform equipped with intelligent algorithms has a promising prospect of replacing manual insulator detection methods. In response to the shortcomings of slow speed and low accuracy in the detection process of insulator defect targets, an improved YOLOv5 insulator defect fault detection method integrating attention mechanism is proposed. This method integrates the SE attention module and CBAM attention module in the YOLOv5s network, and combines the SE attention module with the C3 module in the network structure to enhance the network"s feature extraction ability. The construction of the self built insulator dataset was completed through relevant image processing methods. The k-means++clustering algorithm was used to construct the prior frame of the self built dataset, and the Mosaic-9 data augmentation strategy was introduced, effectively solving the problem of insufficient training data and difficulty in ensuring training effectiveness. Experimental verification shows that the improved detection method improves the accuracy of insulator detection by 9.7% without affecting the detection time, which has certain reference significance for power system inspection methods.