融合结构推理的深度模型输电线路金具及其缺陷检测  

Integrating Structural Reasoning for Deep Model Transmission Line Fittings and Their Defects Detection

作　　者：赵振兵[1,2] 熊静 徐厚东[3] 张凌浩 ZHAO Zhenbing;XIONG Jing;XU Houdong;ZHANG Linghao(Department of Electronic and Communication Engineering,North China Electric Power University,Baoding 071003,China;Hebei Key Laboratory of Power Internet of Things Technology,North China Electric Power University,Baoding 071003,China;Sate Grid Sichuan Electric Company,Chengdu 610000,China;State Grid Sichuan Electric Power Research Institute,Chengdu 610000,China)

机构地区：[1]华北电力大学电子与通信工程系,保定071003 [2]华北电力大学河北省电力物联网技术重点实验室,保定071003 [3]国网四川省电力公司,成都610000 [4]国网四川省电力公司电力科学研究院,成都610000

出　　处：《高电压技术》2023年第8期3346-3353,共8页High Voltage Engineering

基　　金：国家自然科学基金(61871182,U21A20486);四川省科技计划项目(2021YFG0113)。

摘　　要：准确的实现金具及其缺陷的自动化巡检是保证输电线路正常运行的一项重要任务。为了缓解现有各种金具及其缺陷检测方法缺乏上下文信息,导致误检、重检的问题,提出一种基于上下文结构推理的输电线路金具及其缺陷检测方法,在检测模型输出检测结果后加入结构知识,以提高模型的准确率。首先将图片输入到目标检测模型中;之后把检测模型输出的结果送入结构推理模块:将检测结果映射到序列X中,形成一个向量,包括检测框类别、bbox坐标和检测置信度,然后将向量送入双向门控循环单元和自注意力中进行处理,利用输电线路金具及其缺陷的结构知识,提高正确正样本的置信度、降低错误正样本的置信度,最后通过回归器得出最终的输出结果,来达到提高平均精确度的目的。实验结果表明:在加入结构推理模块之后,基线模型的P值均有提高,其中P_(50)相较于基线模型最高提升了6%,为提高输电线路金具及其缺陷检测精度提供了新的思路。It is an important task to ensure the normal operation of transmission lines to realize the automatic inspection of fittings and their defects accurately.To alleviate the problem that various existing metal fittings and their defect detection methods lack context information,which leads to false detection and re-detection,this paper proposes a transmission line metal fittings and their defects detection method based on context structure reasoning,which adds structure knowledge after the detection model outputs the detection results to improve the accuracy of the model.First,the image is input into the object detection model;Then,the output results of the detection model are sent to the structural reasoning module,namely,the de-tection results are mapped to the sequence X to form a vector,including the detection box category,box coordinates,and detection confidence,and then the vector is sent to GRU and self-attention for processing.Using the structural knowledge of the transmission line fittings and their defects,the confidence in the true positive samples is increased,and the confidence of the false positive samples is reduced.Finally,the final output result is obtained through the regressor to achieve the purpose of improving the average precision.The experimental results show that,after adding the structural reasoning module,the P’s value of the baseline model is improved,and the P_(50) is up to 6% higher than that of the baseline model,which provides a new idea for improving the detection accuracy of transmission line fittings and their defects.

关 键 词：金具及其缺陷 深度模型 目标检测 结构推理 自注意力 

分 类 号：TM75[电气工程—电力系统及自动化] TP391.41[自动化与计算机技术—计算机应用技术]