Optimizing Transmission Line Insulator Defect Detection: Lightweight Edge AI on UAVs

Effective transmission line insulator defect detection is crucial for maintaining power grid reliability and safety. Traditional methods, which often rely on ground-based inspections or cloud-based computing, can be slow and inefficient. This paper introduces EdgeIDD, a lightweight solution for real-time UAV-based insulator defect detection leveraging edge computing. By enabling on-site data processing, EdgeIDD reduces the reliance on high-bandwidth communication with central servers and addresses connectivity challenges in remote areas where most power transmission towers are constructed. Built upon YOLOv10, the latest state-of-the-art model with faster speed compared to precedent iterations by removing the post-processing step, EdgeIDD integrates minimalist design principles, replacing the complete backbone with lightweight vanilla blocks and incorporating our novel C2fP block and utilized distribution shifting convolution for accelerated feature fusion. Our model achieves a remarkable 45.2% reduction in total parameters compared to YOLOv10n while maintaining a high mean average precision (mAP) of 97.5% with minimal accuracy loss. With hardware acceleration, EdgeIDD reaches 23.3 FPS on a Raspberry Pi 5 with 7.3x speedup. This advancement in edge computing ensures rapid, reliable insulator defect detection directly from the UAV, even in areas with limited connectivity.