Detection Method of Infrared Overhead Traverse Image Based on Improved YOLOv8

Aiming at the problems of low inspection efficiency, missed detections, and false detections caused by the complex background and large aspect ratio variations of overhead conductors. In this paper, an overhead traverse detection model ES-YOLO with multi-scale features was proposed. Firstly, an ECABlock is designed as the feature extraction module, and a C2fECA enhanced backbone network is constructed to improve the feature extraction capability for overhead conductors. Simultaneously, the SE attention mechanism is embedded to enhance the expressive power of overhead conductor features. Experimental results on a self-made overhead conductor dataset show that compared with YOLOv8, ES-YOLO achieves an average detection precision of 85.6%, which is an improvement of 4.5 percentage points, while the model parameters and computational cost are reduced by 12.1% and 4.2%, respectively, satisfying the requirements of overhead conductor detection.