The reliability of power transmission systems requires accurate detection of defects such as cable damage, insulator loss, and surface rust. These defects have complicated geometric shapes, irregular patterns, and small degradation, making existing detectors for such defects suffer from misdetections and heavy computation burden. To solve these issues, we present an enhanced RT-DETR framework named HyRT-DETR. Our method employs three lightweight modules to strengthen backbone representation, enhance multi-scale feature interaction, and reduce encoder computational overhead. The proposed method is evaluated on a self-constructed dataset for transmission line inspection, which contains three defect classes: cable defect, insulator loss, and rust. Experimental results show that our approach achieves a 4.0% increase in mAP50 over the baseline RT-DETR, with reductions in model size and computation. It surpasses a range of mainstream detection algorithms, validating its effectiveness and potential for deployment in real-time UAV-based power grid inspection systems.