To accurately and quickly evaluate the damage state of link beams after the earthquake, the EB-YOLO-v8n link beam damage identification model was proposed by adding an efficient muti-scale attention module(EMA) and a bidirectional feature pyramid network module structure(BiFPN) to the YOLO-v8n model. Firstly, quasi-static tests of 9 link beams with different parameters were designed and completed. The local damage of the structure under different damage states was recorded and summarized into a dataset. Secondly, Mosaic+Fliplr data enhancement technology was used to enhance the input data to ensure the quality, and a total of 2 612 images were obtained as the dataset. Then, the various damages in the dataset were labeled and input into the model for training. Finally, the effectiveness of each improved module was analyzed through ablation experiments. The results show that the average accuracy of the EB-YOLO-v8 model is higher than that of the other models in this paper. This means the improved model has stronger robustness with basically unchanged parameter quantities.Besides, according to the results of the ablation experiment, the average accuracy of the EB-YOLO-v8n model is 1.2% and 3.8% higher than that of E-YOLO-v8n (introducing an efficient multi-scale attention module) and B-YOLO-v8n (replacing the path aggregation network in the feature pyramid networks (FPN) with a weighted BiFPN), respectively, and it also has a certain advantage in the average recognition time per image.In general, the EB-YOLO-v8n model balances accuracy and speed, fitting the high-precision and speed requirements for post-earthquake damage recognition, which can meet the needs in complex engineering conditions.