Aiming at the problem that a large number of images generated by UAV(unmanned aerial vehicle) during power inspection need to be interpreted manually, which is time-consuming and laborious. With the continuous increase of data, it is difficult to meet the requirements of accuracy and real-time with the naked eye, a deep learning algorithm is proposed to automatically identify the defects of key components of power transmission line, such as insulator, pressure connecting pipe and so on. Firstly, the data sets of insulator and tension line clamp and pressure connecting pipe are established, and then two deep neural networks, Faster R-CNN and YOLOv3, are used to build models with different front-end networks. A total of five models are built. During model training, the pre training model is used to improve the training speed, and the sample amplification technology is used to improve the accuracy of the model. The results show that the Faster R-CNN-Resnet18 model has the best effect, and the average accuracy (map) of all recognition objects on the test set is more than 0.87.