Abstract: Identification and extraction of key elements of transmission channels is an important part of the inspection and management of transmission lines. The deep learning algorithm makes the identification of elements automated and intelligent, but the algorithm requires a lot of knowledge graphs for training. Therefore, we propose a large-scale, multi-scale automated construction method for the knowledge graph of key elements of power transmission channels. This method can improve the sample quality while reducing the workload of manual annotation. The research shows that the accuracy of key elements recognized after training with deep learning algorithm based on the knowledge graph constructed have higher accuracy than that based on traditional object-oriented methods.