广东电力2025，Vol.38Issue(3)：27-36,10.DOI:10.3969/j.issn.1007-290X.2025.03.004
基于数据增强和锚框优化的输电线路金具检测方法
Transmission Line Fitting Detection Method Based on Data Augmentation and Anchor Box Optimization
摘要
Abstract
Corrosion,damages,and dirt on fittings seriously threaten the safe and stable operation of transmission lines.To detect these defects,unmanned aerial vehicles(UAV)are typically used for intelligent inspections.Accurately identifying fittings in high-resolution images captured by drones is crucial for subsequent defect detection.For these images,this paper proposes a fitting recognition method based on data augmentation and anchor box optimization.Firstly,a dataset of typical images containing five categories is constructed using images captured by UAV.These images are then enhanced through rotation,mirroring,and brightness adjustments.Subsequently,an improved Faster R-CNN network is built using ResNet50 with feature pyramid network(FPN)structure,an improved region proposal network(RPN)module,and an ROI_heads module.This networks are trained and tested on the aforementioned dataset.Experimental results show that the proposed improved Faster R-CNN model can effectively detect five types of power fittings,achieving an average precision of 85.84％.This method can assist transmission line inspection and maintenance personnel in the intelligent identification and defect detection of aerial fittings images.关键词
输电线路/金具检测/智能巡检/深度学习/数据增强Key words
transmission lines/fitting detection/intelligent inspection/deep learning/data enhancement分类
动力与电气工程引用本文复制引用
吕旺燕,聂铭,黄丰,罗青红,袁超,李桥..基于数据增强和锚框优化的输电线路金具检测方法[J].广东电力,2025,38(3):27-36,10.基金项目
中国南方电网有限责任公司重点科技项目(GDKJXM20222555(036100KK52222079)) （GDKJXM20222555(036100KK52222079）
