红外成像技术是电气设备在线监测中的一项行之有效的手段，而构建基于红外成像技术的变电站电气设备智能在线监测系统是电气设备监测的发展方向。红外图像分割技术是变电站电气设备智能在线监测系统中智能软件模块的一个重要组成部分，也是系统完成智能监测的一个承上启下的重要环节，通过分割提取的电气设备特征可以为系统后期的智能判断和决策提供依据。 本文以变电站红外监测系统为研究的出发点，以提高系统的自动化、智能化水平为目的，着重以变电站红外图像的分割技术为主要研究内容。本文对当前主流图像分割算法进行了分类、归纳和总结，通过实验了解他们的性能和特点，并对新出现的图像分割方法进行了研究。在此基础上，本文选择了基于形态学分水岭分割算法和模糊聚类分割算法。为了有效准确的检测油枕油位，本文采用改进的分水岭分割算法对油枕红外图像进行分割，改进算法以多尺度形态学梯度图像作为分水岭算法的输入图像，在侧重区域联通性的前提下用大津法二值化梯度图像获得初步标记图，去除伪极小值点后得到最终标记图，用最终标记图中极小值点作为分水岭分割的起点对尺度梯度图像进行分水岭变换得到最终的分割结果。实验证明本文方法能够准确有效地分割出油枕和油位，可以作为变电站在线监测系统的组成部分之一。考虑红外图像信息本身的复杂性和相关性，本文还对模糊核聚类分割算法进行了研究，本文改进的稳健模糊核聚类算法通过加入空间信息来增强算法像素划分的合理性，并提出聚类中心初始化方法来使得聚类迭代尽快收敛至合理划分时的参数值，在分割变压器散热器图像的实验中，本文改进的稳健模糊核聚类算法相比其他模糊聚类方法能够有效分割出散热器故障位置和故障区域。

Infrared imaging technology is a proven means of on-line monitoring for electrical equipment, while the construction of intelligent on-line monitoring system which is based on infrared imaging technology is the direction of developing electrical equipment monitoring system. Infrared image segmentation is an important component of software modules in system and key step to achieve the function of intelligent monitoring. The characteristics of electrical equipment extracted through segmentation can be used as a basis for the system to make intelligent judgments and decisions in the later stage. This paper is based on substations’ infrared monitoring system, aims at enhancing the automation and intelligence level of this system, and therefore pays its main attention to the substations’ infrared image segmentation technology. Firstly, the current image segmentation algorithms of visible image have been classified and summarized in the paper, and the characteristics and the weak point of the algorithms are discussed through experiment comparison. This paper also attaches importance to new image segmentation algorithms. Finally, Watershed segmentation algorithm and fuzzy clustering segmentation algorithm have been selected for segmenting infrared image in this article based on above research work. To be effective and accurate detection of oil level, this paper processed infrared image of the Oil Conservator with the improved watershed segmentation algorithm which has taken the multi-scale morphological gradient image as inputting image of watershed transform. On the premise of focusing on regional connectivity, multi-scale morphological gradient image has been binarization by the method of Otsu to obtain preliminary seed map. After the removal of false minimum marks, we make the exact minimum marks of the final seed map as the starting point of Watershed transform to accomplish image segmentation. Experimental results show that proposed method in this article can separate the oil level of Oil Conservator effectively and accurately. therefore, it can be used as an one of the important component of substation’s on-line monitoring system. Taking into account the internal information complexity and relevance of Infrared image, this paper has conducted a study of Kernel-based fuzzy C-means segmentation algorithm yet. Improved Robust Kernelized Fuzzy C-means(IRKFCM) can enhanced the rationality of segmentation by adding spatial information. And this paper also proposed the clustering center initialization method which can enable the clustering iteration parameters quickly converge to reasonable division parameter values. Experimental results show that IRKFCM algorithm can separate malfunction location of Radiator and calculate the size of malfunction region effectively and accurately compared to other fuzzy clustering method.