基于深度卷积神经网络的变电站设备故障图像分类方法
Deep Convolutional Neural Network-based Method to Classify Fault Images of Substation Equipment
陈昌飞 1康曦 1何振梁 1朱东松 1葛林 1王海1
作者信息
- 1. 国网江西省电力有限公司赣州供电分公司,江西赣州 341000
- 折叠
摘要
为优化变电站设备故障图像分类的效果,提高图像分类的召回率,满足复杂化和多样化的设备故障图像分类需求,开展了基于深度卷积神经网络的变电站设备故障图像分类方法研究.首先,对变电站设备图像进行预处理,减少噪声干扰,达到平滑图像的目的.其次,使用阈值分割算法,对变电站设备图像进行区域分割,为后续故障图像分类奠定良好的基础.在此基础上,利用深度卷积神经网络实现故障图像特征高度提纯,以变电站设备相对温差为依据,判断设备故障缺陷等级,输出故障图像分类结果.实验结果表明,应用所提出的方法后,随着图像样本数量的增、加,其在召回率方面始终保持良好的性能,能够更准确地分类出故障图像样本,具有更好的泛化能力.
Abstract
In order to optimize equipment fault image classification utility,improve classification recall rate,and satisfy image classification requirement of complex and diversified equipment with respect to transforming substation,this work made an attempt at design a deep convolutional neural network-based image classification method.The method entails im-age preprocessing for denoising and smoothing,algorithm-based threshold segmentation,the consequent in-depth feature extraction via deep convolutional neural network,and finally fault level determination and classification according to equip-ment relative temperature difference.The application of the proposed method was demonstrated by experiment capable of maintaining satisfactory recall rate with the increase of image sample amount,more accurate in identifying and classifying fault image samples,and superior in terms of generalization ability.
关键词
深度卷积神经网络/变电站/设备/故障/图像/分类Key words
deep convolutional neural network/transforming substation/equipment/fault/image/classification引用本文复制引用
出版年
2024
NETL
NSTL
万方数据