北京交通大学学报2025，Vol.49Issue(3)：68-78,11.DOI:10.11860/j.issn.1673-0291.20240016
基于改进YOLOv8n的轻量级输电线路异物入侵检测模型
A lightweight foreign object intrusion detection model for transmission lines based on improved YOLOv8n
摘要
Abstract
Aiming at the problems of low accuracy and high model complexity in foreign object detec-tion caused by the large scale variations and variable shape of foreign object targets in the complex envi-ronment of transmission lines,an improved foreign object detection model DLS-YOLOv8n is pro-posed.Firstly,the Bottleneck structure in the C2f module of the backbone network is replaced by the Deformable Convolution Bottleneck module to strengthen the model's feature extraction ability of for-eign object targets with variable shapes and improve the detection accuracy;Secondly,a Light Bi-directional Feature Pyramid Network is proposed to replace the neck network of the original model,which reduces the number of model parameters and computational complexity while improving the detection accuracy of the network on small targets;Thirdly,a parameter free attention mechanism SimAM is added before the model detection head to enhance the model's attention to targets in complex environments.Finally,to validate the performance of the DLS-YOLOv8n model,abla-tion experiments and multiple comparative experiments are conducted on a power line foreign ob-ject dataset.Experimental results show that the proposed algorithm achieves an mAP of 97.1％on the dataset of foreign objects in transmission lines,with a model parameter number of 2.07 M and a computational complexity of 6.9 G.Compared with the original YOLOv8n model,the mAP is increased by 1.6％,and the parameter number and computational complexity are reduced by 31％and 14％,respectively.Compared with one-stage detection models such as(Single Shot MultiBox Detector,SSD),YOLOv5s,and YOLOv7-tiny,the proposed model achieves the highest detec-tion accuracy while maintaining the lowest complexity.The research findings can provide valuable reference and insights for the field of transmission line inspection.关键词
YOLOv8n/异物入侵检测/可变形卷积/特征融合网络/注意力机制Key words
YOLOv8n/intrusion of foreign objects/deformable convolution/feature fusion network/attention mechanism分类
信息技术与安全科学引用本文复制引用
李珅,杜科,李舟演,李宁,熊岑,柳明慧,张云起,秦伦明..基于改进YOLOv8n的轻量级输电线路异物入侵检测模型[J].北京交通大学学报,2025,49(3):68-78,11.基金项目
国家自然科学基金(62073024) （62073024）
国家电网有限公司科技项目(SGSH0000AJJS2310204)National Natural Science Foundation of China(62073024) （SGSH0000AJJS2310204）
State Grid Corporation of China Technology Project(SGSH0000AJJS2310204) （SGSH0000AJJS2310204）
