Ensuring power line reliability is vital to prevent outages; thus, timely maintenance of power line infrastructure is essential. However, traditional inspections are costly and labor-intensive. Recently, UAV-powered vision-based power line inspection solutions have significantly reduced costs and time. However, with varying power line site environments and limited datasets, developing robust power line component identification and detection systems remains a severe challenge. To this end, this study introduces an in-house large-scale power line dataset, PowerCompoDet5. The dataset features 12,770 annotations of power line components and anomalies collected under varied field conditions. We establish a benchmark of 20 state-of-the-art YOLO models, including YOLOv7, YOLOv8, YOLOv9, and YOLOv10, for detecting these components. The detection accuracy across these models ranges from 79.60% by YOLOv7e6 to 89.90% by YOLOv9e. Moreover, YOLOv8n and YOLOv10n show great potential for real-time applications. The results of the comprehensive study presented in this work may be a valuable resource for promoting future research on power line dataset collection, AI-powered power line component detection, and monitoring of power line infrastructure.