摘要
端子排图纸是变电站二次回路施工、检修和运行维护的重要依据。应用计算机视觉技术对其进行电气语义解析存在复杂性和专业性上的双重挑战,使其不可避免地面临难例样本被误解析的窘境,从而阻碍该类方法在真实场景的落地应用。针对端子排图纸电气语义解析存在的不足,提出一类融合图元检测和领域先验的变电站端子排图纸电气语义解析方案。该方案涉及图元检测、文本识别及线帽语义标签解析三个步骤,具体由基于滑动窗冗余切割和改进YOLOv8的多尺度图纸图元检测,以及基于文本识别与领域先验的图纸拓扑匹配与纠错拒识两个算法组成。该方案在基于真实场景构建的图元检测数据集与线帽语义标签解析数据集上分别进行了性能验证,实验结果表明,图元检测算法优于当前主流检测模型,图纸拓扑匹配与纠错拒识算法则在拒识率为10.8%的基础上使得线帽语义标签解析的拟召回率和拟精确率都达到了100%。
Terminal strip drawings are essential for the construction,maintenance,and operation of substation secondary circuits.The application of computer vision technology for their electrical semantic analysis presents dual challenges of complexity and expertise,inevitably leading to the misinterpretation of difficult samples,which hinders the practical application of these methods in real-world scenarios.To address these shortcomings in the electrical semantic analysis of terminal strip drawings,we propose a substation terminal strip drawings electrical semantic analysis scheme that integrates symbol detection and domain knowledge.This scheme involves three steps:symbol detection,text recognition,and wire marker semantic label analysis.It includes multi-scale drawing symbol detection based on sliding window redundancy cutting and an improved YOLOv8,as well as drawing topology matching and error correction rejection algorithms based on text recognition and domain knowledge.Performance validation on a symbol detection dataset and a wire marker semantic label analysis dataset constructed from real-world scenarios shows that the proposed symbol detection algorithm outperforms current mainstream detection models,and the proposed topology matching and error correction rejection algorithm achieves a rejection rate of 10.8%,with both the quasi recall rate and quasi precision rate for wire marker semantic label analysis reaching 100%.
作者
吴勇
葛奕雯
洪文谦
高正霄
戴挈军
李自然
陈蕾
WU Yong;GE Yi-wen;HONG Wen-qian;GAO Zheng-xiao;DAI Qie-jun;LI Zi-ran;CHEN Lei(Jiangsu Power Transmission&Transformation Co.,Ltd.,Nanjing 211106,China;School of Computer Science,Nanjing University of Posts and Telecommunications,Nanjing 210023,China;Jiangsu Key Laboratory of Big Data Security and Intelligent Processing,Nanjing 210023,China)
出处
《计算机技术与发展》
2025年第1期184-191,共8页
Computer Technology and Development
基金
国家重点研发计划项目(2022YFB3303800)
江苏省送变电有限公司科技项目(2023外306)。
关键词
变电站图纸识别
端子排图纸解析
图元检测
文本识别
目标检测
领域先验
substation drawing recognition
terminal strip drawing analysis
symbol detection
text recognition
object detection
domain priors
作者简介
吴勇(1987-),男,高级工程师,研究方向为电力自动化与电力工程建设;通信作者:陈蕾(1975-),男,教授,博导,CCF高级会员(16224S),研究方向为计算机视觉、人工智能与机器学习。
