基于特征增强的输电线路航拍小目标异物检测算法研究

打开文本图片集

Abstract： Foreign objects on transmission lines， due to their small size and weak characteristics， occupy a minimal proportion （0.01%～0.1% of pixels） in drone aerial inspection images， making them prone to being overwhelmed by background noise during downsampling.This results in high false-negative rates and unstable recognition with traditional detection algorithms.To address this，an improved YOLOv8x-based rapid detection model for smallforeign objects on aerial transmission lines is proposed： the backbone network incorporates a P6 multi-scale feature layer and a C3STR module integrated with Swin Transformer to enhance global contextual representation for small targets; a lightweight structure utilizing DWConv and Focus reduces computational load while improving inference speed; and to tackle sample imbalance，an ATFL adaptive threshold focal loss function is introduced to strengthen learning for difficult and smal-sample categories. Experimental results demonstrate a 17.9% improvement in mAP for the enhanced model， with significantly improved detection performance for smalland blurry targets while maintaining real-time capability，providing a deployable technical solution for intelligent transmission line operation and maintenance.

Key words： aerial object detection; deep learning; transmission line inspection; smal target detection; computer vision

0 引言

输电线路的安全稳定运行是电力系统供电可靠性的根本保障。（剩余11699字）