In the long-term operation of power grids, foreign objects on transmission lines can affect the safe and stable operation of transmission lines and even lead to large-scale power outages, which bring incalculable losses to social and economic development. In this paper, based on the characteristics of the acquired transmission line images, we construct a transmission line foreign object dataset, improve the loss calculation and network structure of the basic Faster RCNN based on the RCNN network model target detection method, and detect foreign objects on transmission lines, such as plastic film, kites, bird nests, ice, etc., by training the model. The final experimental results show that compared with the traditional object recognition method, the improved Faster R-CNN not only overcomes the instability of manually extracted features, but also has the ability to resist complex environmental interference and improves the accuracy of foreign object detection on transmission lines.