TLDRT-DETR: Adaptive Upsampling and Dual-Activation Attention for Real-Time Transmission Line Defect Detection

Defects in transmission line components pose serious threats to the safety and reliability of power systems. However, accurate detection remains challenging due to large scale variations, complex outdoor backgrounds, and the coexistence of subtle and prominent defect patterns. To address these challenges, we propose TLDRT-DETR, an improved real-time detection framework based on RT-DETR. Specifically, we design an Adaptive Attention Dynamic Upsampling (AADU) module to replace conventional upsampling in cross-scale feature fusion, enabling content-adaptive feature reconstruction and better preservation of multi-scale structural information. In addition, a Dual-activation Spatial and Channel Synergistic Attention (DualActSCSA) module is introduced into high-level feature fusion to enhance defect feature discriminability and suppress background interference. Experimental results show that the proposed method achieves 90.5% Precision, 85.4% mAP@50, and 55.7% mAP@50:95, outperforming the RT-DETR baseline by 2.2%, 1.7%, and 1.8%, respectively. These results demonstrate that TLDRT-DETR provides more accurate and robust defect detection in complex transmission line inspection environments.