基于深度学习的红外图像中劣化绝缘子片的分割方法

Faulty insulator segmentation method in infrared image based on deep learning

近年来,红外成像法日渐成为绝缘子在线检测重要技术手段,而当前对于电气设备红外图像的分析仍大多依靠人工经验,缺乏智能化。传统的红外图像中劣化绝缘子的分割方法需要复杂的图像预处理和人工提取目标特征,因而存在着泛化能力差、复杂背景下识别精度低等缺陷。基于以上问题,文中提出一种基于深度学习的红外图像中劣化绝缘子的分割方法。搭建实验平台,获得绝缘子串在有其他设备和热源干扰情况的复杂背景下的红外图像。为了保证红外图像的多样性,拍摄在多种污秽等级下进行,选取了多种阻值的劣化绝缘子片放置在了多种位置,并不断改变了拍摄的角度。构建多尺度特征融合的方法实现全卷积神经网络(FCN)的3种子模型架构:FCN-32s、FCN-16s和FCN-8s,使用随机梯度下降的训练方法对模型参数进行训练,最终实现了红外图像中的劣化绝缘子片的自主分割提取。通过对三种子模型训练测试对比后,得出以下结果:FCN-8s模型为劣化绝缘子片分割提取最佳模型,对于测试数据集像素分类准确率为89.23%。结果表明文中所描述的智能化红外劣化绝缘子片分割方法,为绝缘子和其他高压电气设备的红外在线监测诊断打下了坚实基础。

As an important means of on-line detection of insulators,infrared imaging technology has matured in recent years.But the current analysis of infrared images still mostly depends on manual experience,which is lack of intelligence.The traditional methods of faulty insulator segmentation in infrared images requires complex image preprocessing and manual extraction of target features,which leads to the disadvantages of poor generalization capability and low recognition accuracy in complex contexts.Based on the above problems,this paper proposes a method of faulty insulator segmentation in infrared images based on deep learning.Firstly,an experimental platform was set up to obtain infrared images of insulator strings under complex background.In order to ensure the diversity of infrared images,shooting was conducted in a variety of contamination levels,shooting angles,and backgrounds.Meanwhile,insulator pieces with various degrees of deterioration were selected and placed in various positions.Then,a multi-scale approach is built to achieve the three sub-models of fully convolutional network(FCN)architecture,including FCN-32 s,FCN-16 s,and FCN-8 s.The stochastic gradient descent method was applied to the training of this end-to-end model,and finally,the autonomous division of faulty insulators in infrared images is achieved.After training and comparing the three sub-models,the following results are obtained that the FCN-8 s model is the best model for faulty insulator segmentation,and the accuracy rate for verifying the test data set is 89.23%.The results show that the intelligent infrared faulty insulator segmentation method described in this paper has a good segmentation effect,and lays a solid foundation for intelligent infrared on-line monitoring and diagnosis of other high-voltage electrical equipment.