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 (6): 292-298,304. doi: 10.19678/j.issn.1000-3428.0059872

• 开发研究与工程应用 • 上一篇    下一篇

基于改进Faster RCNN的输电线路航拍绝缘子检测

易继禹, 陈慈发, 龚国强   

  1. 三峡大学 计算机与信息学院, 湖北 宜昌 443000
  • 收稿日期:2020-10-29 修回日期:2021-01-03 发布日期:2021-01-08
  • 作者简介:易继禹(1997-),男,硕士研究生,主研方向为深度学习、图像识别;陈慈发,教授;龚国强,副教授。

Aerial Insulator Detection of Transmission Line Based on Improved Faster RCNN

YI Jiyu, CHEN Cifa, GONG Guoqiang   

  1. School of Computer and Information, China Three Gorges University, Yichang, Hubei 443000, China
  • Received:2020-10-29 Revised:2021-01-03 Published:2021-01-08
  • Contact: 国家重点研发计划(2016YFB0800403)。 E-mail:yjy463268889@163.com

摘要: 为提高航拍图像中输电线路绝缘子的检测准确性,提出一种改进的Faster RCNN网络模型。在原始Faster RCNN网络模型上运用多尺度训练,同时根据绝缘子自身特性调整滑动窗口产生的候选区域比例,并引入检测困难样本的对手生成策略,实现不同尺寸及部分遮挡输电线路绝缘子的准确检测。实验结果表明,改进的Faster RCNN网络模型相比原始Faster RCNN网络模型的检测精确度提升了4.33个百分点,能更准确地检测出目标绝缘子。

关键词: 电力巡检, 绝缘子, 目标检测, 多尺度训练, 对手生成策略

Abstract: To improve the detection accuracy of transmission line insulators in aerial images, this paper proposes an improved Faster RCNN model.The method applies multi-scale training to the original Faster RCNN model, and adjusts the proportion of the candidate regions generated by sliding window according to the characteristics of insulators.In addition, an opponent generation strategy for detecting difficult samples is introduced to realize accurate detection of insulators for different sizes and partial shielded transmission lines.Experimental results show that the accuracy of the improved Faster RCNN model is 4.33 percentage points higher than that of the original Faster RCNN model, proving that the model can detect the target insulators more accurately.

Key words: power inspection, insulator, object detection, multiscale training, opponent generation strategy

中图分类号: