Abstract: In order to solve the shortcomings problem of traditional transmission line defect detection methods, this paper proposes an intelligent identification method of transmission line defects based on reinforcement learning and Transformer. First, the deterministic networking(DetNet) with a large receptive field is used to extract the features of the inspection defect image of the transmission line, and then the deep Q-network(DQN) is used to screen out the important areas containing foreground information. Secondly, based on bilinear attention mechanism, the feature vectors of the background region are compressed by projection, so that the fused feature vectors focus on the target region. Finally, the reliability evaluation index is defined for the uncertain defect detection results, the adaptive adjustment mechanism of the coding level of the Transformer network is constructed, and the Transformer model library with different coding levels is established to obtain the multi-level differential features of multi-modal defects. The integrated detection results are obtained by using the soft-NMS to improve the robustness of the identification model. Through experimental research on aerial images of transmission line defects, the average detection accuracy of this method is 89.7%, which has better detection accuracy and generalization ability than other algorithms.