基于遮挡关系推理的输电线路图像金具检测

PDF下载HTML

Title:

Fittings detection in transmission line images with occlusion relation inference

作者:

戚银城^1,2, 赵席彬¹, 耿劭锋¹, 张薇¹, 赵振兵^1,2, 吕斌³

1. 华北电力大学 电子与通信工程系，河北 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，河北 保定 071003;
3. 国网浙江杭州市萧山区供电有限公司，浙江 杭州 310000

Author(s):

QI Yincheng^1,2, ZHAO Xibin¹, GENG Shaofeng¹, ZHANG Wei¹, ZHAO Zhenbing^1,2, LYU Bin³

1. Department of Electronic and Communication Engineering, North China Electric Power University, Baoding 071003, China;
2. Hebei Key Laboratory of Power Internet of Things Technology, North China Electric Power University, Baoding 071003, China;
3. State Grid Zhejiang Hangzhou Xiaoshan Power Supply Co., Ltd., Hangzhou 310000, China

关键词:

输电线路; 金具; 遮挡关系描述; 结构推理; 超快速区域卷积神经网络; 目标检测; 门控循环单元; 图

Keywords:

transmission line; fittings; occlusion relationship description; structure inference; Faster R-CNN; object detection; gated recurrent unit; graph

分类号:

TP18; TM726

DOI:

10.11992/tis.202108036

文献标志码:

2022-10-09

摘要:

实现输电线路图像典型金具的精准检测是进行其缺陷检测的前提。针对通用目标检测模型对密集分布、遮挡严重的金具检测精度较低、易出现漏检等问题，提出了一种结合金具间遮挡结构信息和场景关联信息的典型金具检测方法。基于经典的Faster R-CNN模型提取金具特征作为节点，提取整张图像特征作为金具场景关联信息，学习金具标注框间相交区域信息作为金具遮挡关系信息，并采用图同时建模金具特征、场景关联信息和遮挡关系信息，通过门控循环单元信息传递机制构建结构推理模块完成金具类别和位置的联合推理检测。为了验证所提方法的有效性，选取了8类存在遮挡连接关系的金具进行实验，其中，原始Faster R-CNN模型的mAP值为81.30%，改进模型的mAP值为84.15%。实验结果表明，本文方法一定程度上提高了遮挡严重金具的检测精度，为后续的金具故障诊断奠定良好的基础。

Abstract:

The accurate detection of typical fittings in transmission line images is the premise of fault detection. This study proposes a typical fittings detection method that combines occlusion structure information and scene association information to address the problems of low detection accuracy and missed detection of the common target detection model, such as dense distribution and serious occlusion of the fittings. Based on the classical Faster R-CNN detection model, the method extracts features of the entire image of fittings as scene association information, learns the intersecting area information between the marking frames as occlusion structure information, uses a graph to model the feature of fittings, scene-related information and occlusion structure information, and constructs a structure reasoning module through the information transmission mechanism of the gated recirculating unit to complete the joint inference detection of the category and position of fittings. Experiments with eight types of fittings with occlusion relationships are chosen to validate the effectiveness of the proposed method. The Faster R-CNN model shows an mAP value of 81.30%, while the proposed model has an mAP value of 84.15%. The experiments show that the proposed method can improve the detection accuracy of serious occlusion fittings to some extent and that it has laid a good foundation for the subsequent fault diagnosis of the fittings.