External Defect Recognition of Lightning Arresters Based on an Improved YOLOv10 Model

To address the external defect detection challenges of lightning arresters in power systems, this paper proposes a novel visual detection method, Ar-YOLOv10, based on an improved YOLOv10 model for identifying external defects in lightning arresters. First, we constructed and publicly released a comprehensive dataset of lightning arrester external defects to provide essential data support for related research. The dataset used in this study is publicly available through the Roboflow platform at https://app.roboflow.com/godlizy/arrester-external-defect-dataset/1. Subsequently, we designed comparative experiments to evaluate the performance of four real-time object detection models: Ar-YOLOv10, YOLOv10, RT-DETR, and SSD. Experimental results demonstrate that the Ar-YOLOv10 model achieved a mean Average Precision of 82.12% on the test set, significantly outperforming the other three models. Moreover, Ar-YOLOvl0 exhibited excellent performance across critical metrics, including Precision, Recall, and detection speed. The research findings validate the effectiveness and practicality of the proposed method in real-time detection of external defects in lightning arresters.