基于迁移学习方法的架空配电线路无人机精准定位及拍摄技术研究  

Research on UAV accurate localization and photographing technology of overhead distribution network line based on migration learning method

作　　者：周鼎 何涛 周杰 黄亮 杨国林 ZHOU Ding;HE Tao;ZHOU Jie;HUANG Liang;YANG Guolin(Yongchuan Power Supply Branch of State Grid Chongqing Electric Power Company,Chongqing 402160,China;State Key Laboratory of Power Transmission and Distribution Equipment and System Safety and New Technology(Chongqing University),Chongqing 400044,China)

机构地区：[1]国网重庆市电力公司永川供电分公司,重庆402160 [2]输配电装备及系统安全与新技术国家重点实验室(重庆大学),重庆400044

出　　处：《供用电》2024年第10期68-74,83,共8页Distribution & Utilization

基　　金：国家自然科学基金项目(52077018)。

摘　　要：使用无人机(unmanned aerial vehicle,UAV)代替传统的徒步巡检配电线路更加符合目前新型电力系统的发展需求。针对传统滤波器和梯度存在不足,以及线路环境嘈杂,因而无法精准检测配电线路的问题,提出了一种基于迁移学习方法的UAV架空配电线路精准定位和拍摄技术。首先介绍了该学习方法及其过程中用图像重新训练成适用于UAV巡检的点实例网络(point instance network,PINet)车道线检测模型。然后通过多次训练PINet车道线检测模型,提出了在其前后分别添加旋转比较器模块和后处理模块的方法,得到了一种新的配电线路检测技术。最后在手持相机拍摄图像、UAV拍摄图像、山地线路数据集、城区线路数据集4个数据集中进行了训练验证,结果表明该方法均达到了合格的准确率、误报率和低漏检率,尤其在城区线路数据集上的效果超越了现有其他方法。The use of unmanned aerial vehicle(UAV)instead of the traditional foot inspection of distribution network lines is more in line with the current development needs of the new power system.In response to the backwardness of the traditional filter and gradient's technology,coupled with the noisy line environment,which can not accurately detect the distribution lines,this paper proposes a migration learning method based on the UAV overhead distribution line accurate localization and photographing technology.The learning method and its process of retraining with images into a PINet vehicle road detection model suitable for UAV inspection are first introduced.Then by training the PINet lane line detection model for many times,the method of adding the rotary comparator module and the post-processing module before and after it respectively is proposed,and a new distribution network line detection technique is obtained.Finally,the training validation is carried out on four datasets,namely,handheld camera captured images,UAV captured images,power line dataset of mountain and power line dataset of urban scene,and the results show that the method achieves qualified accuracy,false alarm rate,and low leakage detection rate,and the results especially on the power line dataset of urban dataset outperform the existing methods.

关 键 词：配电网 无人机 迁移学习 线路检测 PINet 

分 类 号：TM72[电气工程—电力系统及自动化]