摘要
为实现配网带电作业机器人在自动化搭接引流线过程中对引流线的精确分割与定位,解决高压强电磁干扰、复杂背景及引流线细长、弱纹理等带来的识别与定位难题,本文提出了一种基于DeepLabV3+的引流线分割方法,并结合双目相机实现引流线的定位。首先,对DeepLabV3+模型进行了3方面改进:使用MobileNetV3替换原始主干网络,明显降低了模型的计算复杂度;在空洞空间金字塔池化(atrous spatial pyramid pooling,ASPP)模块中插入卷积块注意力机制(convolutional block attention module,CBAM),增强了模型对引流线关键区域的关注能力;在模型输出端增加形态学处理,通过腐蚀和膨胀方法减小了噪声对分割结果的干扰。其次,利用双目相机的立体视觉技术,结合分割结果提取引流线的关键点,通过立体匹配算法实现定位。试验结果表明,本文方法在机器人的引流线分割任务中的精度较高(M_(IoU)达到84.48%),在三维定位中的定位误差小于2 cm。本文研究为配网带电作业机器人的智能化操作提供了可靠的技术支持,为后续引流线自动识别提供了参考。
In order to realize the accurate segmentation and positioning of the jumper wires during the process of automatic lapping operation of the jumper wires by the distribution network with power operation robot,and to address the identification and positioning problems caused by high voltage strong electromagnetic interference,complex background and slender and weak texture of the jumper wires,in this paper,a method based on DeepLabV3+is proposed to segment jumper wires,and opsition jumper wires combined with binocular camera.Firstly,three improvements are made for the DeepLabV3+model:the original backbone network is replaced by MobileNetV3,which significantly reduces the computational complexity of the model;convolutional block attention module(CBAM)is inserted into the atrous spatial pyramid pooling(ASPP)module,which enhances the model’s ability to focus on the key areas of the jumper wires;morphological processing is added to the model output,which reduces the interference of noise on the segmentation results by corrosion and expansion.Secondly,the stereo vision technology of binocular camera is utilized to extract the key points of the jumper wire in combination with the segmentation results,and the localization is realized by stereo matching algorithm.The experimental results show that the method in this paper achieves high accuracy(M_(IoU)reaches 84.48%)in the task of segmentation of jumper wire,and realizes the localization error less than 2 cm in 3D localization.This research provides reliable technical support for the the intelligent operation of distribution network with power operation robots,and lays a foundation for the accuration identification of jumper wires.
作者
陈乐
樊绍胜
彭音音
CHEN Le;FAN Shaosheng;PENG Yinyin(School of Electrical&Information Engineering,Changsha University of Science and Technology,Changsha 410114,China)
出处
《电力学报》
2025年第5期329-338,共10页
Journal of Electric Power
基金
国家自然科学基金(62473065)
国家重点基础研究发展计划项目(973计划)(2006CB200303,2006CB2003056)。
关键词
配网带电作业机器人
引流线分割
DeepLabV3+
双目相机
定位
distribution network with power operation robot
jumper wire segmentation
DeepLabV3+
binocular camera
localization
