在供电网络发展极为迅速和网架结构日趋合理的今天，国家对电力系统的智能化和供电可靠性的要求越来越高，要求能够自动快速隔离电气设备的故障并恢复健全区域的供电，能对电力系统进行在线监测和安全预警，及时发现缺陷并采取措施消除隐患。红外检测技术作为一种行之有效的故障检测手段，已得到广泛应用。将红外检测技术与图像监控系统相结合大大提高了电力系统故障检测水平，但是目前的检测手段依然需要人工诊断，无法达到实时准确。本文针对变电站电气设备故障问题，利用数字图像处理技术来实现无人值守变电站电气设备故障的自动检测。 首先，通过分析变电站红外图像的特点，对红外图像中存在的噪声进行了归纳总结，并实验验证了几种常见的去噪处理方法，在此基础上进行了改进，提出了具有高信噪比的中值与均值相结合的去噪方法，既能消除噪声的扩散影响，又能使图像区域内的灰度值更为平滑，完成了图像的预处理。 其次，针对红外图像低信噪比、低对比度的特点，在研究分析传统的图像分割方法的基础上，提出了分水岭与K均值聚类相结合的分割方法，并实例验证了该方法不仅能够克服分水岭方法的过分割现象，也克服了聚类方法对噪声的敏感性，能够有效、准确地提取目标区域。 最后，根据变电站红外图像中像素空间信息的特点，本文提出了基于图像灰度信息的目标检测方法和基于像素灰度信息的目标检测方法，实验表明：本文算法不仅能够检测出变电站电气设备中故障所在，且实验的实验值与参考值的误差处在2%的可忽略范围内，验证了本文方法的可行性和准确性。该方法结合红外遥视和巡检系统，可应用于无人值守变电站，实现电气设备从传统的“事后维修”、“定期预防维修”方式向“状态维修和预知维修”的方向发展。

Nowadays, with the rapid development of Electricity Networks and rational network structure, the country requires increasingly high demanding for intelligent power system and the reliability of power supply, such as: requiring automatic rapid isolation for electrical equipment failure and restore the power supply of the whole region; requiring the functions of online monitoring and safety warning for the electricity system, which would detect the defects timely and take measures to eliminate hidden dangers. Infrared detection technology has been widely used as an effective means. The combination of infrared detection technology and image monitoring system greatly improve the level of power system fault detection, but current detection methods still require manual diagnosis, unable to achieve real-time and accurate. To target on the problem of electrical substation equipment failure, the automatic detection of electrical equipment failure in unmanned substation is achieved by using digital image processing technology. Firstly, the noises in the infrared images are summarized by analyzing the characteristics of the infrared image of the substation, and experiments several common de-noising methods. On this basis, then proposed de-noising method with a high signal-to-noise ratio of the combination of median and mean, not only can eliminate the diffusion effects of noise, but also make the gray value within the image area more smoothly, completed the process of image preprocessing. Secondly, considering the characteristics of the low signal to noise ratio and low contrast for the infrared images, on the basis of analysis of the traditional image segmentation methods, a new segmentation method which combined the Watershed transformation and K-means clustering segmentation method is proposed and verified. The results show that this method not only be able to overcome the phenomenon of over-segmentation of the watershed method, but also to overcome the noise sensitivity of the clustering method, and it could extract the target region effectively and accurately. Finally, according to the characteristics of the substation infrared image pixel spatial information, Target detection method based on image gray level information and Target detection method based on pixel spatial information, which of electrical substation equipment failure are proposed. Experiments show that: The algorithm can not only detect the fault location of substation electrical equipment, but also make the experimental error between the actual value and the reference value in the 2% negligible range, which could verify the feasibility and accuracy of the proposed method. The combination of this method and the system of infrared remote viewing and inspection can be applied to unmanned substations, to make the electrical equipment to develop from the traditional “corrective maintenance” and “preventive maintenance on a regular basis” approach to the direction of “the state of maintenance and predictive maintenance”.