作者投稿和查稿 主编审稿 专家审稿 编委审稿 远程编辑

计算机工程 ›› 2021, Vol. 47 ›› Issue (8): 301-307,314. doi: 10.19678/j.issn.1000-3428.0058448

• 开发研究与工程应用 • 上一篇    下一篇

基于先验知识的航拍绝缘子检测方法

宋万潼1, 李冰锋1, 费树岷2   

  1. 1. 河南理工大学 电气工程与自动化学院, 河南 焦作 454000;
    2. 东南大学 自动化学院, 南京 210096
  • 收稿日期:2020-05-17 修回日期:2020-07-16 发布日期:2020-07-28
  • 作者简介:宋万潼(1994-),男,硕士研究生,主研方向为目标检测;李冰锋,讲师、博士;费树岷,教授、博士。
  • 基金资助:
    河南省高等学校重点科研项目(16B120001);河南理工大学博士基金(722103,722001,722208)。

Prior Knowledge-Based Insulator Detection Method Using Aerial Images

SONG Wantong1, LI Bingfeng1, FEI Shumin2   

  1. 1. School of Electrical Engineering and Automation, Henan Polytechnic University, Jiaozuo, Henan 454000, China;
    2. School of Automation, Southeast University, Nanjing 210096, China
  • Received:2020-05-17 Revised:2020-07-16 Published:2020-07-28

摘要: 在架空输电线路中对带电状态的绝缘子进行检测和故障定位,对保证电网可靠运行具有重大意义。基于无人机平台提出一种复杂背景条件下的绝缘子检测算法。在检测算法的特征提取层引入注意力机制模块以获取更多的绝缘子特征信息,同时利用航拍图像中绝缘子的先验知识,结合K均值聚类算法改进目标候选框的生成模式。在此基础上,通过将中心损失引入绝缘子检测目标函数以增强训练过程中绝缘子类内特征的内聚性。实验结果表明,相对Faster R-CNN检测算法,在绝缘子检测数据集上Faster R-CNN改进算法检测精度提高4%以上。

关键词: 先验知识, 绝缘子检测, K均值聚类, 注意力机制, 候选框, 中心损失

Abstract: The detection and failure location of live insulators in overhead transmission lines is important to the reliable operation of power grid. Based on the Unmanned Aerial Vehicle(UAV) platform, this paper proposes an algorithm for insulator detection under complex conditions. An attention mechanism module is introduced into the feature extraction layer of the detection algorithm to obtain more information of insulator features. Then based on the prior knowledge of the insulators in aerial images, The K-means clustering algorithm is used to improve the generation mode of the object candidate box. On this basis, the center loss is introduced into the objective function of insulator detection in order to enhance the cohesion of in-class features of insulators during the training. The experimental results show that compared with the Faster R-CNN algorithm, the proposed Faster R-CNN algorithm improves the accuracy by more than 4% on the insulator detection data set. a

Key words: prior knowledge, insulator detection, K-means clustering, attention mechanism, candidate box, center loss

中图分类号: