用于配网线缆识别和定位的多传感器引导系统
Multi-sensor Guidance System for Identification and Localization of Distribution Network Cables
杨淼 1夏骏 1李金亮 1王邹俊 1李哲2
作者信息
- 1. 国网湖南省电力有限公司电力科学研究院,湖南 长沙 410007
- 2. 湖南大学 电气与信息工程学院,湖南 长沙 410082
- 折叠
摘要
不停电配网作业视场条件复杂,传统机器视觉方法存在配网线缆识别定位精度与速度无法满足要求等问题.为实现智能化配网作业中配网线缆的快速准确识别与定位,该文设计了多传感器视觉引导系统以及搭载的配网线缆分割算法.首先,面向配网作业需求进行了针对性的传感器选型,提出了一套使用可见光相机+激光雷达(RGB+Lidar)对末端执行机构进行引导的系统,以解决传统配网检修人工作业方式劳动强度大、工作效率低等问题.其次,针对传统机器视觉方法参数量大、推理速度慢的问题,将深度可分离卷积引入轻量化的图像分割模型的设计,输入部分增加了雷达点云提供稀疏的深度信息,并引入了直线注意力模块以进一步提高精度.最后,经过配网作业中采集的数据集进行图像分割模型的测试,验证了其速度和精度能够满足不停电配网环境下的识别定位需求.
Abstract
Due to the complex conditions of the power-on distribution network operation,the traditional machine vision methods have problems that the recognition and positioning accuracy and speed of distribution network cables cannot meet the requirements.In order to realize the fast and accurate identification and positioning of distribution network cables in intelligent distribution network operation,we design a multi-sensor vision guidance system and a distribution network cable segmentation algorithm.Firstly,the sensor selection is carried out for the needs of distribution network operation,and a system using RGB+Lidar to guide the end execution mechanism is proposed to solve the problems of high labor intensity and low efficiency of the traditional manual operation mode of distribution network maintenance.Secondly,to address the problems of large parameter count and slow inference speed of traditional machine vision methods,depth separable convolution is introduced into the design of the lightweight image segmentation model.Moreover,the radar point cloud is added to the input part to provide sparse depth information.We also introduce the straight line attention module to further improve accuracy.Finally,the image segmentation model is tested with the dataset collected in the distribution network operation,and its speed and accuracy are verified to be able to meet the identification and localization requirements in the environment of the power-on distribution network.
关键词
深度学习/图像处理/距离测量/多传感器融合/配网线缆Key words
deep learning/image processing/range finding/multi-sensor fusion/distribution network cables引用本文复制引用
基金项目
国网湖南省电力有限公司科技项目(5216AS310001)
出版年
2024
国家科技期刊平台
NSTL
万方数据