YOLOv10-MsA: Attention-Augmented Real-Time Insulator Defect Detection from UAV Imagery
The reliable operation of transmission systems depends on early detection of defects within high- voltage insulators because it helps stop power outages. Scalable inspection remains a challenge since UAV-based imagery and deep learning generate recent promising solutions. The goal of this study is to build an accurate and time-efficient insulator defect detection system through the YOLOv10 architecture integration of Manhattan Self-Attention (MsA), which strengthens spatial feature detection and increases robustness during evaluations in complex aerial inspection scenarios. The designers implemented the MsA modules into the backbone and neck sections of YOLOv10 to develop YOLOv10-MsA as their novel detection model. The model relied on 5,000 annotated insulator images acquired by unmanned aerial vehicles throughout various defect classes during training and evaluation. Standard object detection metrics consisting of mAP@0.5, mAP@0.5:0.95, precision, recall, F1-score, and inference speed evaluated the performance of the model. The YOLOv10-MsA system reached an mAP@0.5 performance of 93.1% and an F1-score of 91.9% at an inference speed of 79 FPS, which surpasses YOLOv8, YOLOv9, and baseline YOLOv10. The model demonstrated its best performance at detecting various small and hard-to-detect defects such as chipping and contamination. The application of MsA in detection systems resulted in better accuracy while preserving real-time operation, according to related model assessment. The proposed YOLOv10-MsA serves as a powerful deployable system for UAV-based insulator inspection because it achieves both high accuracy and fast operation. The method establishes conditions for real-time smart infrastructure observation with attention-augmented frameworks for object detection.