摘要
为解决目前绝缘子污秽等级检测方法操作繁琐、漏判率高的问题,提出了一种基于红外热像处理的瓷绝缘子污秽等级检测方法。首先,将现场拍摄的绝缘子红外热像图进行二步法去噪;然后,针对绝缘子红外热像的颜色特征,采用RGB图像中的R分量图以及改进的OTSU分割算法获取二值图像,再对二值图像进行两次数学形态学修正,得到绝缘子积污区域。利用统计学方法提取积污区域R通道的5个特征分量,并对BP神经网络进行训练,建立绝缘子污秽等级判别网络。选取5个污秽等级的XP-70瓷绝缘子红外热像图共500组作为试验样本,对该系统进行反复测试,总识别精度达到91%以上。
With the popularization of high or extra-high voltage power transmission and the continuous improvement of grade of wire and voltage, some power grids frequently have pollution flashover accidents. Therefore, it is very important to effectively and accurately obtain the insulator contamination grade for the safe operation of power grid.Currently, major detecting methods of insulator contamination grades include equivalent salt deposit density method,leakage current method, frequency spectrum method, and so on. Although above mentioned methods are feasible,basically, all of them mainly focus on the treatment of single insulator without paying attention to the entire group of insulators' heating features and appearance characteristics of the heating zone. Therefore, according to the heating features of polluted insulators, a new detecting method of porcelain insulator contamination grade, which was based on color image treatment, was proposed under the purpose of simplifying the procedures as well as reducing the misdetection ratio of the present detecting method. It meant that the assessment of samples' contamination grades would be performed under a color- image- based treatment on the obtained insulator infrared thermogram. Firstly, the infrared thermograms, which were taken at the scene, were denoised by two- step method(Gaussian noise was denoised by the wavelet adaptive diffusion after the impulse noise was denoised by the median filtering method) to obtain the filtered insulator thermogram. Secondly, they were segmented based on the differences of the component R, G, and B respectively shown by the normal insulators, the broken insulators, and the corresponding infrared thermograms.Although the traditional OTSU segmentation algorithm is simple and effective, it is not suitable to apply on the image segmentation without clear bimodal histogram. Therefore, the presented paper completed the image threshold segmentation by combining the characteristics of wavelet transform and OTSU. The modified OTSU was adaptive to choose the optimal threshold value, and improved the obtained image quality after the segmentation. However, the binary images obtained from the segmentation still had lots of noise points and cavities. Therefore, the mathematical morphology of the obtained binary image was modified twice in order to obtain a polluted area of the insulator. Thirdly,5 feature components, including ratio of area to perimeter, average value, extreme value, standard deviation as well as major- minor axis ratio of the smallest outer ellipse, were extracted from the R channel of the polluted area by taking advantage of a statistical method, and the back propagation(BP) neural network was trained. During the training, the transfer function of the hidden layers was the logarithmic transfer function, logsig. The neural transfer function of the output layers was the linear activation function, purelin, which regarded image characteristic values as the input of the network, and the salt deposit density as the output that would be transformed to the corresponding contamination grade.Consequently, a detecting model of contamination grades was constructed on the basis of the color image. Finally, the running conditions of the insulators in the Qitaihe Power Supply Company were tested, among which the relative humidity, one of the shooting conditions, was at 78%- 91% in the experiment. However, the temperature of the experiment was adjusted as 10-30 ℃ because the infrared thermogram was hardly affected by the shooting temperature.The contamination grade of each insulator was recorded according to the regulations of GB/T 5582- 4993 when each infrared thermogram was shot. A total of 500 groups of infrared thermograms were obtained from XP- 70 porcelain insulator with 0,Ⅰ,Ⅱ, Ⅲ and Ⅳ polluted grade respectively, which were selected as experimental samples. This system was tested repeatedly, and the overall detecting precision reached above 91%. The experimental result shows that the insulator contamination grades are more accurate when the appearance characteristics of polluted insulators and the features of infrared thermogram are combined during the detection. It can also be noted that such a method can be served as the foundation of the detection of insulator contamination grades in the complicated outdoor environment.
出处
《农业工程学报》
EI
CAS
CSCD
北大核心
2016年第15期175-181,F0003,共8页
Transactions of the Chinese Society of Agricultural Engineering
基金
教育部春晖计划资助项目(Z2012074)
关键词
瓷
神经网络
算法
绝缘子
污秽等级
红外热像
porcelain
neural networks
algorithms
insulator
contamination grades
infrared image