摘要
本文基于可见光机巡图像技术提出了电力设备杂草智能识别方法。以可见光无人机巡检图像内存在的杂草特征,与卷积神经网络相结合,切实解决图像内电力设备周围杂草识别等相关问题。以针对机巡图像增广数据,并引进区域生成网络,提取图像中的基础信息,再面向图像提取固定数量候选框的图像特征,与优化后图像分类网络相衔接,从而构成整体卷积神经网络模型,以智能识别电力设备周围存在的杂草。通过实验分析,结果表明,基于可见光机巡图像技术的电力设备杂草智能识别方法准确率高达98.24%,而600×600的图像检测只需耗费约0.25s,在保障高准确率的基础上,实现了高效快速识别。
In this paper,an intelligent weed identification method for power equipment is proposed based on the visual inspection image technology.Based on the characteristics of weeds in the inspection image of visible UAV,combined with convolution neural network,the problem of weed identification around the power equipment in the image can be solved effectively.In order to enlarge the data of the patrol image,and introduce the region generating network,extract the basic information in the image,then extract the image features of a fixed number of candidate frames from the image,and connect with the optimized image classification network,thus forming the overall convolution neural network model to intelligently identify the weeds around the power equipment.Through the experimental analysis,the results show that the accuracy of the intelligent weed identification method based on the visible light machine patrol image technology is as high as 98.24%,while the 600×600 image detection only costs about 0.25s,on the basis of ensuring the high accuracy,it realizes the efficient and fast identification.
作者
毛先胤
马晓红
王雪晨
MAO Xian-yin;MA Xiao-hong;WANG Xue-chen(Electric Power Research Institute of Guizhou Power Grid Co.,Ltd.,Guiyang 550002 China;Suzhou Ansoft Information Technology Co.,Ltd.,Taicang 215400 China)
出处
《自动化技术与应用》
2021年第11期118-121,126,共5页
Techniques of Automation and Applications
基金
贵州电网有限责任公司科技项目(编号0666002020030101GY00001)。
关键词
可见光
机巡图像
电力设备
杂草识别
智能
visible light
patrol image
power equipment
weed identification
intelligence