基于高光谱成像技术的500 kV运行绝缘子污秽识别及可视化研究

Research on Recognition and Visualization of 500 kV AC Insulator Contamination Based on Hyperspectral Imaging Technology

电网绝缘子数量多,污秽成分复杂,传统污秽测试方法耗费大量人力物力,为快速、有效地辨识其污秽程度,本文提出一种基于高光谱成像技术的线路绝缘子污秽等级识别技术。首先,利用高光谱成像仪对不同污秽等级的84支运行绝缘子样品进行图谱信息采集,进行黑白校正;其次,获取感兴趣区域(region of interest,ROI)的反射率光谱曲线,使用SG平滑预处理方法削弱噪声干扰,通过竞争性算法(CARS)提取波谱特征;同时,采用HSI分量法分析现场绝缘子不同污秽等级的图像特征,提取图像饱和度(Saturation)分量特征;结合光谱数据与图像特征参量,建立基于极限学习机(Extreme Learning Machine,ELM)的污秽等级识别模型,识别准确率可达86.1%,获取绝缘子高光谱图像各像素点的图谱信息并使用已建立的污秽等级识别模型进行分类,可实现绝缘子污秽等级的图像可视化。

There are a large number of power grid insulators and the pollution components are complex.The traditional pollution tests use insulator as a unit for pollution detection,and it is difficult to distinguish the non-uniformity of pollution.In this paper,a recognition technology of line insulator pollution level and non-uniformity based on hyperspectral imaging technology is proposed.Firstly,the hyperspectral imager was used to collect the spectrum information of 84 insulator samples with different pollution levels,and the region of uneven pollution distribution was obtained after black-and-white correction The SG smoothing pretreatment is used to reduce the interference,and the competitive algorithm(CARS)is used to extract the spectral features;then the HSI component method is used to analyze the pollution image features,and the image saturation component is extracted;finally,the extreme learning machine(ELM)is established to classify the pollution grades of all the positions of insulators,and the image visualization of uneven distribution of insulator pollution is realized.The recognition accuracy is 86.1%.