An Insulator Location and Defect Detection Method Based on Improved YOLOv8

Ensuring the integrity of insulators is critical for the reliability and safety of power transmission systems. To address the need for efficient and real-time inspection of insulator defects on power lines, this paper introduces an advanced defect detection model built upon the YOLOv8 architecture. The model incorporates a novel C2f-Faster-EMA module that modifies the original C2f module used in YOLOv8's backbone for feature extraction. This adaptation employs FasterNet to reduce the model's parameter count, while incorporating an EMA-based attention mechanism to enhance detection accuracy. Additionally, we replace the conventional PANet structure with a BiFPN-P feature fusion module to improve the extraction of shallow features, which is crucial for detecting small-target defects in insulators. Further refinements include the implementation of the Inner-IoU concept to augment the MPDIoU loss function, thus improving the model's ability to learn from challenging samples. Experimental results demonstrate that the proposed Insulator-YOLOv8s model achieves a superior performance over existing mainstream algorithms, with a mean Average Precision (mAP) of 91.5% at an IoU threshold of 0.5. The model is characterized by a parameter count of 5.66M and computational requirements of 21.1 GFLOPs, achieving detection speeds up to 113 frames per second. These enhancements enable the proposed model to identify insulator defects swiftly and with high accuracy, thereby contributing significantly to the safety and maintenance of power transmission infrastructure.