基于深度学习的变电站表计检测与识别  

Substation Meter Detection and Recognition Based on Deep Learning

作　　者：窦永亮 刘呈弟 李少旭 胡文亮 梁强春 DOU Yongliang;LIU Chengdi;LI Shaoxu;HU Wenliang;LIANG Qiangchun

机构地区：[1]中核汇能(甘肃)能源有限公司,甘肃兰州735000

出　　处：《电力系统装备》2025年第2期87-90,共4页Electric Power System Equipment

摘　　要：智能电网的快速发展对变电站的表计识别提出了高效率和高精度的要求。传统人工抄表方式效率低且受环境影响较大,特别是在风沙环境下识别率显著下降,不利于变电站的自动化管理。为了解决上述问题,文章提出了一种基于深度学习的变电站表计检测与识别方法,通过图像处理与特征提取模型,在YOLOv5s框架中完成表计识别,并结合自动校正与破损检测模块提升识别的准确性与鲁棒性。试验结果表明,该方法在风沙环境中显著优于传统模式识别方法,识别准确率达到92%,并有效减少环境干扰因素的影响。The rapid development of smart grids has raised demands for high efficiency and precision in meter recognition at substations.Traditional manual meter reading methods are inefficient and highly susceptible to environmental factors,particularly in windy and dusty conditions,where recognition rates significantly decrease,hindering the automated management of substations.To address these issues,this paper proposes a meter recognition method based on deep learning.Through image processing and feature extraction models,meter recognition is accomplished within the YOLOv5s framework,and automatic correction and damage detection modules are integrated to enhance recognition accuracy and robustness.Experimental results demonstrate that this method significantly outperforms traditional pattern recognition methods in windy and dusty environments,achieving a recognition accuracy rate of 92%and effectively mitigating the impact of environmental interference factors.

关 键 词：表计识别 深度学习 YOLOv5s 自动校正 破损检测 

分 类 号：TM63[电气工程—电力系统及自动化]