摘要
输电杆塔是输电线路的重要组成部分,采用人工或无人机等手段对杆塔相关目标进行周期性巡检至关重要。为解决目前待检测杆塔类型不同、距离远、尺寸小、图像畸变等问题,本文提出一种基于深度学习框架的杆塔相关目标检测方法。该方法在YOLOv5s网络的基础上使用FReLU激活函数代替SiLU激活函数,进而改善了对杆塔相关目标的图像识别准确率;为解决上述问题,采用了对小目标、低分辨率友好的SPD-Conv模块;同时使用对目标特征分批次处理的SPPCSPC空间金字塔池化,进一步提升了杆塔目标的检测精度。实验结果表明,改进后的YOLOv5s-FSS网络相比原YOLOv5s网络,其平均精度(mAP)提升2.4%,查准率(Precision)提升0.3%,查全率(Recall)提升1%,目标检测性能提升效果显著,能够有效提升输电杆塔巡检效率。
Transmission towers are critical components of power transmission lines,and periodic inspections of tower-related targets using methods such as manual inspection or unmanned aerial vehicle are of utmost importance.To solve problems such as different tower types,long distances,small sizes,and image distortions encountered in current tower inspections,this paper proposes a deep learning-based tower-related target detection method.The method builds upon the YOLOv5s network,utilizing the FReLU activation function instead of the SiLU activation function,which improves the accuracy of tower-related target recognition.To tackle issues with small targets and low resolutions,the SPD-Conv module is introduced to process the targets,making the detection of small objects more effective.Additionally,the SPPCSPC spatial pyramid pooling is employed to handle target features in batches,further enhancing the detection precision of tower-related objects.Experimental results demonstrate that the improved YOLOv5s-FSS network outperforms the original YOLOv5s network,with an increase of 2.4%in mean average precision(mAP),0.3%in precision,and 1%in recall.This significant enhancement in target detection performance efficiently improves the efficiency of transmission tower inspections.
作者
朱辉
李海涛
刘岳鑫
赵玮
钱骁
刘禹涵
高明阳
ZHU Hui;LI Haitao;LIU Yuexin;ZHAO Wei;QIAN Xiao;LIU Yuhan;GAO Mingyang(State Grid Changzhou Power Supply Company,Jiangsu Electric Power Co.,Ltd.,Changzhou 213000,Jiangsu,China)
出处
《电力大数据》
2023年第5期62-72,共11页
Power Systems and Big Data
关键词
深度学习
输电杆塔
信息识别
激活函数
小目标
空间金字塔池化
deep learning
transmission towers
information recognition
activation function
small goals
spatial pyramid pooling
作者简介
朱辉(1974),男,本科,高级工程师,主要从事输电、电缆线路运维、检修管理等方面工作。
