Anomaly Detection of Transmission Line Large Metal Based on EGFPN-YOLO and UAVs

Abnormal large fittings in transmission lines are a common fault in the safe operation of power systems. In this paper, a detection model named EGFPN-YOLO is proposed. The model combines the latest YOLOv8n algorithm, with particular improvements in the detection head, designing an efficient and lightweight detection head called EfficientHead. Additionally, inspired by the RepGFPN method, the Neck part of the model is improved by introducing the CSPStage structure to enhance the integration capability of different level features. These innovations not only improve the detection accuracy but also reduce the computational and parameter complexity of the network, achieving a dual enhancement in speed and simplicity. Experimental results show that compared to the baseline network, our model achieves an increase in detection speed from 131.58 frames per second to 769.23 frames per second, an improvement in mAP50 detection accuracy from 68.3% to 79.4%, and an increase in mAP50-95 from 35.1% to 42.4%. These significant performance improvements make our method a more efficient solution for detecting abnormally large fittings in transmission lines.