基于支持向量机的高压绝缘子污秽等级红外热像检测

Infrared Thermal Image Detecting of High Voltage Insulator Contamination Grades Based on Support Vector Machine

提出了一种红外热像处理与支持向量机多值分类器相结合的新方法对高压绝缘子污秽等级 进行检测。采用基于梯度信息的自适应平滑滤波方法对原始图像进行了滤波处理;利用OTSU图 像分割方法对滤波后的图像进行分割,获取绝缘子盘面区域,并从绝缘子盘面区域提取了最高温 度、最高温度与最低温度的比值、盘面温度的标准偏差、部分最高温度像素点个数与目标总像素点 个数的比值共4个反映污秽程度的红外特征量;设计了支持向量机多值分类器对绝缘子污秽等级 进行分级。试验结果表明,文中所选取的绝缘子红外特征量可有效表征绝缘子的污秽等级,所采用 的支持向量机多值分类器是一个小样本、高效率的分类器,所提出的绝缘子污秽等级检测新方法是 可行的。

This paper presents a new method that integrates infrared thermal image technique with support vector machine （SVM） classifiers to check the contamination grades of high voltage insulators. Firstly, a self-adaptive smooth filter based on gradient information is used to remove the noise of the original image. Secondly, the OTSU segmentation method is adopted to segment the target area from the above filtered image. Four features are then extracted from the surface of the insulator, viz. maximum temperature, contrast ratio between maximum and minimum temperatures, and standard deviation of insulator surface temperatures, and ratio between the top 10 ~//0 brightness pixels and the total pixels. Finally, a multi-class SVM is designed for insulator contamination grades detecting. Experimental results indicate that four insulator infrared features selected in this paper can represent the contamination degree of an insulator effectively and the multi-class SVM is of high efficiency with merely small samples. Thus it further shows the new approach proposed is feasible.