Abstract: The zero resistance insulator is one of the important reasons causing grounded fault of transmission and distribution network. The key of detecting zero resistance insulators by infrared thermal imaging is to obtain the optimal and appropriate infrared thermal image features. An infrared thermal image feature selection method based on Logistic regression analysis is proposed. The infrared thermal image is denoised with median filtering and wavelet adaptive diffusion, and the contrast of the image is enhanced by means of gray stretching. The image is segmented to binary image with two-dimensional maximum entropy threshold. After binary image filling, the minimum enclosing rectangle around the disc surface of insulator string is intercepted automatically to obtain rectangle image of the insulator string area, through gray-gradient co-occurrence matrix, 13 texture features of rectangle area are extracted, and 14 characteristic parameters consisting of pollution class and texture features are screened by Logistic regression analysis, with 7 parameters among them having significant influence on the classification result. The experiments show that the proposed selection method is simple, and can effectively screen characteristic parameters and eliminate redundant data.

Key words: zero resistance insulator, infrared thermal image, Logistic regression analysis, minimum enclosing rectangle, texture feature

摘要： 零值绝缘子是造成输配电网络对地短路故障的重要原因之一。利用红外热像技术检测零值绝缘子的关键在于获取最优且适当的红外热像特征。提出了一种Logistic回归分析筛选零值绝缘子红外热像特征的方法。利用中值滤波及小波自适应扩散法进行红外热像去噪和灰度拉伸法增强图像对比度；采用二维最大熵阈值分割形成二值图，经过二值图像填充处理，自动截取绝缘子串区域最小外接矩形，得到绝缘子串区域矩形图像，通过灰度-梯度共生矩阵提取矩形区域13个纹理特征参数；应用Logistic回归分析对污秽等级和纹理特征组成的14个特征参数进行筛选，得出其中7个参数对分类结果有显著性影响。实验表明，该方法实现简单，能有效筛选特征参数，剔除冗余数据。

关键词: 零值绝缘子, 红外热像, Logistic回归分析, 最小外接矩形, 纹理特征