Transformer-Based Detection of Hidden Hazards in Power Transmission Lines for UAV Inspection

Accurate and real-time detection of hidden hazards in power transmission lines is critical for ensuring the safety and reliability of modern power grids, particularly in Unmanned Aerial Vehicle (UAV)-based inspection scenarios. In this paper, we propose STDISNet, a novel Transformer-based object detection framework tailored for complex transmission environments. The model integrates a Swin Transformer backbone to enhance multi-scale semantic representation through hierarchical self-attention, a Dual-Branch Feature Fusion Convolution (DFFConv) module to enhance contextual feature extraction while maintaining low computational overhead compared to traditional convolutional methods, and a SIoU-based loss function to optimize bounding box regression by incorporating geometric constraints. Extensive experiments on a publicly available transmission hazard dataset demonstrate that STDISNet achieves state-of-the-art performance in both detection accuracy and inference speed. Qualitative visualizations further confirm its robustness in challenging scenarios such as occlusion, background clutter, and low-contrast objects. The proposed approach provides a reliable and efficient solution for UAV-assisted intelligent inspection of power transmission systems.