首页 | 本学科首页   官方微博 | 高级检索  
     检索      

复杂场景下的电力线自动提取
引用本文:赵延峰,胡耀垓,王先培,赵乐.复杂场景下的电力线自动提取[J].测绘通报,2021,0(8):1-6.
作者姓名:赵延峰  胡耀垓  王先培  赵乐
作者单位:武汉大学电子信息学院,湖北 武汉430070
基金项目:国家自然科学基金(51707135)
摘    要:实现复杂场景下航拍图像的端到端电力线提取是电力线无人机检测的关键。本文通过分析基于深度学习的电力线实例分割实现存在的问题,提出了一种基于改进Mask R-CNN的电力线自动提取算法。首先,根据电力线的线性特征,提出一种线性IoU计算方法,改进Mask R-CNN原有的IoU计算,提高电力线提取性能;然后,将改进后的网络在电力线数据集上进行训练,得到电力线粗提取结果;最后,通过线段编组拟合算法,对粗提取结果进行聚类拟合,以解决电力线断裂和误检的问题。试验结果表明,所提方法能从环境复杂的无人机航拍图像中较为准确地提取完整的电力线。

关 键 词:电力线提取  无人机  Mask  R-CNN  线性IoU  线段编组拟合
收稿时间:2020-10-14
修稿时间:2021-06-22

Automatic power line extraction algorithm in complex scene
ZHAO Yanfeng,HU Yaogai,WANG Xianpei,ZHAO Le.Automatic power line extraction algorithm in complex scene[J].Bulletin of Surveying and Mapping,2021,0(8):1-6.
Authors:ZHAO Yanfeng  HU Yaogai  WANG Xianpei  ZHAO Le
Institution:School of Electronic Information, Wuhan University, Wuhan 430070, China
Abstract:End-to-end power line extraction from aerial images in complex scenes is the key to power line detection by UAVs. In this paper, an automatic power line extraction algorithm based on improved Mask R-CNN is proposed by analyzing the problems existing in the realization of power line instance segmentation based on deep learning. First of all, according to the linear characteristic of power line, this paper puts forward a kind of improved Mask R linear IoU calculation method, a CNN original IoU calculation, improve the performance of power line extraction. Then the improved network on the power line data set for training, power line crude extraction results. Finally through a line marshalling fitting method, the result of the crude extract clustering fitting, to solve the problem of power line breakage and misdetection. Experimental results show that the proposed method can accurately extract the complete power lines from UAV aerial images with complex environment.
Keywords:power line extraction  UAV  Mask R-CNN  linear IoU  line segment group fitting  
本文献已被 CNKI 万方数据 等数据库收录!
点击此处可从《测绘通报》浏览原始摘要信息
点击此处可从《测绘通报》下载免费的PDF全文
设为首页 | 免责声明 | 关于勤云 | 加入收藏

Copyright©北京勤云科技发展有限公司  京ICP备09084417号