基于前景提取与多特征决策融合的电力线触树放电痕迹识别  

Discharge trace recognition of power line contact trees based on foreground extraction and multi-feature decision fusion

作　　者：邵楠 王连辉 梁栋[1] 邹国锋[1] SHAO Nan;WANG Lianhui;LIANG Dong;ZOU Guofeng(School of Electrical and Electronic Engineering,Shandong University of Technology,Zibo 255049,China;State Grid Fujian Electric Power Co.,Ltd.,Fuzhou 350003,China)

机构地区：[1]山东理工大学电气与电子工程学院,淄博255049 [2]国网福建省电力有限公司,福州350003

出　　处：《集成电路与嵌入式系统》2024年第8期35-43,共9页Integrated Circuits and Embedded Systems

基　　金：国家电网有限公司总部管理科技项目(No.5500202221138A11ZN)。

摘　　要：针对电力线路智能化巡检中难以准确辨识电力线触碰树木遗留痕迹,无法为事故防治和责任划分提供可靠依据的难题,提出一种基于前景区域提取与多特征决策融合的电力线放电痕迹识别方法。首先,搭建中压线路触树放电实验平台,采集导线表面放电遗留痕迹,构建图像集。然后,提出融合纹理稀疏性与对数变换的去阴影算法,消除阴影区干扰;进一步提出融合全局阈值和自适应阈值分割的目标初始框选取策略,解决了GrabCut算法中初始框无法自动确定的问题,实现复杂背景下导线痕迹的自动精准分割。最后,提取导线的纹理和颜色特征,通过多SVM决策融合实现导线痕迹识别。实验结果表明,所提方法平均识别准确率达到88.39%,证明了所提识别方法的有效性。During the intelligent inspection of power lines,it is difficult to accurately identify the traces caused by power lines touching trees,and cannot provide a reliable basis for accident prevention and responsibility division.To solve these problems,the power line discharge trace identification method based on foreground area extraction and multi-feature decision fusion is proposed.Firstly,the experiment platform of contact tree discharge of medium voltage line is built,and the surface discharge trace is collected and the image set is constructed.Then,a de-shading algorithm combining texture sparsity and logarithmic transformation is proposed to eliminate the interference of shadow region.Furthermore,a target initial frame selection strategy combining global threshold and adaptive threshold segmentation is proposed,to address the problem that initial frame cannot be determined automatically in GrabCut algorithm.It realizes automatic and accurate segmentation of line trace under complex backgrounds.Finally,the texture and color features are extracted,and the power line trace recognition is realized by multi-SVM decision fusion.The experimental results show an average recognition accuracy of 88.39%,proving the effectiveness of the proposed method.

关 键 词：树线放电故障 痕迹识别 GRABCUT 灰度共生矩阵 颜色矩 决策融合 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]