基于改进YOLOv5的电力线路安全检测与异常识别方法  

Power line safety detection and anomaly identificationmethod based on improved YOLOv5

作　　者：杨云皓 韩国政[1] 朱国防[2] YANG Yunhao;HAN Guozheng;ZHU Guofang(School of Information and Automation,Qilu University of Technology(Shandong Academy of Sciences),Jinan 250353,China;School of Electrical Engineering,Shandong University,Jinan 250100,China)

机构地区：[1]齐鲁工业大学(山东省科学院)信息与自动化学院,山东济南250353 [2]山东大学电气学院,山东济南250100

出　　处：《齐鲁工业大学学报》2025年第2期1-7,共7页Journal of Qilu University of Technology

基　　金：国家自然科学基金(U23B20122)。

摘　　要：在电力系统中,电力线路承载着输送电能的重要任务,安全可靠的电力线路对电力系统的稳定发展有着重要的意义。针对利用深度学习算法对电力线路中的悬挂物、塔吊等目标检测准确率低、实时性差的问题,提出基于改进YOLOv5的电力线路安全检测与异常识别方法。该方法基于YOLOv5算法,在网络的部分C3模块中引入ECA注意力机制,增强网络的特征提取能力;在主干网络中的池化层之后增加Fast-RFB模块,提高检测速度与准确性;使用解耦头部替代原始网络的耦合检测头部,提高检测精度;最后将原始模型的CIoU损失函数替换成Wise-IoU损失函数,减少训练过程的损失。仿真实验表明,改进后的YOLOv5算法在电力线路数据集上的P mA 0.5与P mA 0.5:0.95分别为92.2%和56.5%,分别超出了YOLOv5原始模型10.3%和7.3%,检测速度为83帧/s,满足实际环境中对电力线路安全检测与异常识别的要求。In the power system,power lines carry the important task of conveying electric energy,and safe and reliable power lines are of great significance to the solid development of the power system.Aiming at the problems of low accuracy and poor real-time detection of hanging objects,tower cranes and other targets in power lines using deep learning algorithms,a power line safety detection and anomaly identification method based on improved YOLOv5 is proposed.The method is based on the YOLOv5s algorithm,which introduces the ECA attention mechanism in part of the C3 module of the network to enhance the feature extraction capability of the network;adds the Fast-RFB module to enhance detection speed and accuracy after the pooling layer in the backbone network;uses the decoupled head to replace the coupled detection head of the original network to improve the detection accuracy;and finally,replaces the CIoU loss function of the original model with the Wise-IoU loss function of the original model to reduce the loss of training process.Simulation experiments show that the P mA 0.5 and P mA 0.5:0.95 of the improved YOLOv5s algorithm on the power line dataset are 92.2%and 56.5%,respectively,which exceeds the original YOLOv5 model by 10.3%and 7.3%,respectively,and the detection speed is 83 frames/s,which meets the requirements of safety detection and anomaly identification of power lines in the real environment.

关 键 词：电力线路 深度学习 注意力模块 解耦头部 

分 类 号：TM75[电气工程—电力系统及自动化] TP391.4[自动化与计算机技术—计算机应用技术]