基于改进轻量化YOLOv8n的输电线路绝缘子故障检测模型  

A Fault Detection Model for Transmission Line Insulators Based on Improved Lightweight YOLOv8n

作　　者：陈鑫 唐东峰[1] 江拼 唐钊 黄新凯 CHEN Xin;TANG Dongfeng;JIANG Pin;TANG Zhao;HUANG Xinkai(School of Information and Electrical Engineering,Hunan University of Science and Technology,Xiangtan 411201,China)

机构地区：[1]湖南科技大学信息与电气工程学院,湖南湘潭411201

出　　处：《湖南科技大学学报(自然科学版)》2025年第1期60-68,共9页Journal of Hunan University of Science And Technology:Natural Science Edition

摘　　要：针对航拍巡检高压输电线路上绝缘子目标易受复杂背景和部分遮挡影响,造成传统算法难以准确检测的问题,提出了一种基于改进YOLOv8n的轻量化检测模型.模型整体由主干特征提取网络、颈部特征融合网络、头部检测网络等3个部分构成.利用C2f_Star替换主干网络和颈部网络中C2f模块,很大程度上缩减了模型大小;在头部网络部分引入细节增强卷积(DEConv)捕捉更细微的故障特征,增强了对局部特征的表达能力;通过组归一化(GN)技术提高了模型在数据上的泛化能力;在定位分支采用可变形卷积(DCNV2),以适应目标的几何变形,实现更精确的定位.模型还通过任务分解与特征融合策略,加强分类与定位任务间的交互,进一步提升检测性能.利用无人机巡检图像制作数据集,将改进算法与4种经典目标检测算法进行比较实验,结果表明:该模型的平均检测精度可以达到98.5%,每张图片的检测时间为0.01 s,兼具了检测的准确率和轻量化.Aiming at the problem that insulator targets on high-voltage transmission lines are difficult to be accurately detected by traditional algorithms due to complex backgrounds and partial occlusions in aerial inspection,this study proposes a lightweight detection model based on the improved YOLOv8n.The model is composed of three parts,i.e.the backbone feature extraction network,the neck feature fusion network,and the head detection network.By replacing the C2f module in the backbone network and the neck network with C2f_Star,the model size is significantly reduced.In the head network,detail enhancement convolution(DEConv)is introduced to capture more subtle fault features and enhance the expression ability of local features.Group normalization(GN)technology is used to improve the model's generalization ability on data.Deformable convolution(DCNV2)is adopted in the localization branch to adapt to the geometric deformation of the target and achieve more accurate localization.The model also enhances the interaction between classification and localization tasks through task decomposition and feature fusion strategies,further improving the detection performance.A dataset is created using aerial inspection images of unmanned aerial vehicles,and the improved algorithm is compared with four classic object detection algorithms.The results show that the average detection accuracy of the model can reach 98.5%,and the detection time for each image is 0.01 s,achieving a balance between detection accuracy and lightweight.

关 键 词：输电线路 绝缘子故障检测 YOLOv8n 轻量化模型 深度学习 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]