基于YOLO的输电线路鸟巢检测网络结构改进研究  

Improvement of Bird’s Nest Detection Network Structure of Transmission Lines Based on YOLO

作　　者：徐鹏雷 杨文刚[2] XU Penglei;YANG Wengang(Huizhou Power Supply Company,Guangdong Power Grid Co.,Ltd.,Huizhou 516000,China;North China Electric Power University,Baoding 071000,China)

机构地区：[1]广东电网有限责任公司惠州供电局,广东惠州516000 [2]华北电力大学,河北保定071000

出　　处：《智慧电力》2024年第4期54-61,共8页Smart Power

基　　金：国家自然科学基金资助项目(52201214);中央高校基本科研业务费专项资金资助项目(2022MS096)。

摘　　要：为提高电力系统的安全运行水平,针对输电线路上的鸟巢识别问题,提出基于YOLO的输电线路鸟巢检测网络。首先通过构建GhostNet模块搭建骨干网络,并优化了特征层提取方式;随后通过改进特征金字塔连接层,并结合PANet结构构建了瓶颈网络的特征金字塔,最终搭建了YOLO-NEST网络。构建并扩充数据集进行训练,将提出的网络与其他目标检测算法进行对比,结果表明所提网络在进行输电线路的鸟巢检测时效率更高。In order to improve the safe operation level of power system,aiming at the problem of bird's nest identification on transmission lines,a bird’s nest detection network based on YOLO is proposed.Firstly,the backbone network is built by constructing GhostNet module,and the feature layer extraction method is optimized.Then,by improving the feature pyramid connection layer and combining PANet structure,the feature pyramid of bottleneck network is constructed,and finally the YOLO-NEST network is built.A data set is constructed and expanded for training.The proposed network is compared with other target detection algorithms,it is concluded that the proposed network is more efficient in bird’s nest detection of transmission lines.

关 键 词：鸟巢 目标检测 网络优化 检测平台 YOLO 

分 类 号：TM755[电气工程—电力系统及自动化]