An Improved Lightweight RT-DETR Algorithm for High-Voltage Transmission Line Inspection

To address technical challenges in drone-based high-voltage transmission line inspection scenarios—including multi-scale diversity of equipment targets, severe complex background interference, and constrained edge computing resources—this paper proposes a lightweight real-time detection model, RT-DETR-MBD. Firstly, the MobileNetV4 architecture is restructured to extract core features. Secondly, the BiFPN multi-scale feature fusion mechanism is optimized within the neck network. Finally, the innovative introduction of a Dynamic Branch Balancing Structure (DBBS) and SE attention mechanism enhances RepC3, achieving precise fusion and adaptive processing of critical information and multi-scale features. Experiments on a proprietary drone-captured infrared dataset of power equipment demonstrate that the improved model maintains 94.3% detection precision while reducing model parameters to 33.6MB—a 46% decrease compared to the baseline RT-DETR-l model. Inference speed increases by 36%, meeting real-time inspection requirements. This research provides an efficient solution for lightweight object detection in complex high-voltage transmission line inspection scenarios, demonstrating significant engineering application value.