In existing transmission line defect detection systems, the issues of low detection accuracy and small target sizes are particularly prominent. To address these challenges, researchers have proposed an improved algorithm based on TD-YOLOv8. This solution first employs a multi-scale SSFPN fusion network to tackle the problems of low detection accuracy and multi-scale targets—this network enhances detection accuracy by integrating deep feature maps with small-target feature maps. Secondly, a DynamicATSS label assignment strategy is designed to increase the number of high-quality samples, thereby improving detection accuracy and accelerating model convergence. Finally, the EIOU Loss is selected as the regression loss function, which not only enhances precision but also speeds up the convergence of regression detection. Experimental results demonstrate that the improved TD-YOLO model reduces the parameter count by 0.1M and computational complexity by 0.4 GFLOPS. Compared to the original YOLOv8 algorithm, its mAP50 metric improves by 3.9%, and the mAP metric increases by 1.7%, achieving significantly higher detection accuracy for transmission line damage in harsh environments.