The safety and reliability of transmission lines are very important to the stable operation of power system, and defects of insulators and wires directly impact transmission efficiency and power grid safety. The traditional manual inspection method is inefficient and easily influenced by human factors. A deep learning (DL) model is constructed to automatically analyze image data of insulators and conductors, enabling the quick and accurate identification of defects. In the study, in collaboration with power companies, images and video data of transmission lines in different geographical locations and climatic conditions were collected. The main defect types of insulators and conductors were identified, and a dataset containing about 4,650 images was constructed. The model's performance is enhanced through data preprocessing and augmentation, and ResNet-50 is selected as the core model for training. The experimental results show that the model based on DL is significantly superior to the traditional method in recognition accuracy, and the accuracy reaches 92.50%, which shows the effectiveness and potential of DL technology in power system safety monitoring and maintenance. Despite the challenges of strong data dependence, high demand for computing resources, and poor interpretation, this study provides a new technical path for fault identification of transmission lines and lays a foundation for future optimization and application.