MEF-YOLOv8-Based Foreign Object Detection Method in Transmission Lines Under Complex Environments

The steady operation of transmission lines is crucial for maintaining the stability of power systems. Addressing the issue of low accuracy in identifying common foreign objects such as nests, garbage, kites, and balloons on transmission lines under sophisticated environments, this paper presents a MEF-YOLOv8 model based on the YOLOv8n baseline. Firstly, the Mosaic-9 data augmentation method is adopted at the input stage of the training phase to improve the network's generalization capability. Secondly, the Efficient Multi-Scale Attention (EMA) module is integrated into the backbone of the YOLOv8 network enabling pixel-level attention in feature maps and enhancing the model's ability to capture feature information of foreign objects of different scales. Lastly, to boost the accuracy of foreign object detection in intricate environments, the Focal-EIOU loss function has been introduced to supersede the initial CIOU loss function. The empirical findings indicate that the proposed approach has achieved a 99% [email protected] score in identifying typical foreign bodies on transmission lines, representing a 4.9% enhancement compared to the original YOLOv8 algorithm.