利用红外图像特征和径向基概率神经网络识别不同湿度条件下绝缘子的污秽等级

Contamination Grades Recognition of Insulators Under Different Humidity Using Infrared Image Features and RBPNN

提出一种利用污秽绝缘子红外图像特征和径向基概率神经网络(RBPNN)来检测不同湿度条件下自然污秽绝缘子污秽等级的新方法。采用修正后的阿尔法滤波器和基于波谷的图像分割方法对绝缘子红外图像进行预处理。提取了不同湿度条件下的图像背景(周围环境)的平均温度、绝缘子盘面区域的最高温度、绝缘子盘面区域的平均温度、绝缘子盘面温度分布的方差值作为反映污秽等级的4个特征量。通过RBPNN建立了湿度及污秽特征与污秽等级之间的映射关系,并利用训练好的RBPNN识别绝缘子污秽等级;另外提出一种梯度算法与随机性方法相结合的算法来确定RBPNN的隐中心、宽度控制参数及权值矩阵。实验结果证明该方法能有效识别不同湿度条件下绝缘子的污秽等级。

A new method is presented, using infrared image features and radial basis probabilistic neural network （RBPNN） for insulator contamination grades detection of natural contaminated insulators under different humidity. An amended alpha filter and an image segmentation method based on the histogram trough of the insulator image are adopted to preprocess the insulator infrared image, Experiments are designed under different humidity, four features which can represent the contamination grades of the insulator are extracted from the segmented image, they are the mean temperature of the image background, the highest temperature of the insulator surface, the mean temperature of the insulator surface, the temperature variance of the insulator surface, RBPNN is designed to map the relation between the infrared image features of the insulator under different humidity and the contamination grades; then RBPNN is trained to recognize the contamination grades. A method integrated gradient algorithm and random algorithm is applied to decide the centers of hidden nodes, the width control parameters and the weights matrix of the RBPNN. Experiments results indicate this method is an effective approach for the detection of the insulator contamination grades.