绝缘子是接触网支撑及悬挂装置的关键部件，起着悬挂装置中带电部分与绝缘部分的电气隔离及对悬挂装置的支撑作用等等。但是由于工作时间长，工作环境恶劣，绝缘子容易出现破损及夹杂异物的状态，轻则影响绝缘性能，重则造成跳闸，严重影响列车运营安全。
    本文首先概要描述了机器学习理论技术，多核并行技术发展历史及OpenMP并行开发方案，并对本文相关图像处理与识别技术展开介绍，包括高斯滤波、基于Canny算子的边缘提取技术、基于Hough变换的直线提取技术以及介绍多种图像相似度对比技术等。在此基础上，本文结合铁路局绝缘子异常检测需要，使用用例图的方式作出了系统需求分析和总体设计框架。在数据库设计方面，本文利用实体图及设计表格等方式对数据库设计进行说明。
    随后本文介绍了绝缘子异常检测系统的五个子模块设计与实现，分别为图像预处理模块、机器学习模块、异常识别模块、人工精检模块和多核并行检测模块。在图像预处理模块，系统通过随机概率统计的方式过滤过度曝光或曝光不足的巡检图片，通过提取直线信息对重复拍摄的图片加以剔除，并进行图像增强处理；在机器学习模块，系统通过对巡检样例图片提取特征进行机器学习建模，建立绝缘子结构级联分类器，识别绝缘子所在图片区域；在异常识别模块，系统基于Gabor滤波判断绝缘子偏斜角度，并通过灰度统计的方式判断绝缘子是否存在缺陷异常；在人工精检模块，本文通过时序图的方式展开设计，阐述了人工精检系统确定绝缘子缺陷异常的过程；在多核并行检测实现模块，本文结合OpenMP多核编程框架阐述多核技术在本系统各功能模块的应用过程。
    本文最后介绍了系统开发及测试环境，并分析系统各功能模块在不同分辨率图片、不同CPU核心数、不同应用场景等基础上的运行效果。实验证明本系统能在不同CPU核心的计算机上高效运行，并能以较高的处理速度和检测效率对绝缘子异常进行自动化检测，精检页面人机交互效果良好。

 

Insulators are important parts of Catenary suspension and supporting devices, which perform well in isolating live part and insulation part. However, due to the long working hours and poor working environment, insulators are prone to damaged and be contained with foreign matters. It might affect the performance of insulation, even cause tripping, and it makes great challenge to train operation safety. 

Above all, this paper introduces the machine learning theory and technology, the development history of multi-core parallel technology including OpenMP development scheme. In addition, this paper introduces the related image processing and image recognition technology, including Gauss filter, edge extraction technology based on Canny operator, line extraction technology based on Hough Transform and several image similarity comparing technology. On this basis, combined with the need of insulator anomaly detection of the Railway Bureau, this paper use User Case Diagram to make the system requirements analysis and the overall design framework. This paper describes the design of database by means of the Entity Diagram and tables.

Subsequently, this paper introduces the design and implementation of the five sub modules of the anomaly detection system. The five sub modules are image preprocessing module, machine learning module, anomaly detecting module, manual accurate detecting module and multi-core parallel detecting module. In image preprocessing module, this system filters the picture of excessive exposure or lack of exposure, and the picture of repeat shooting by extracting line information. Later, the system enhances the image. In machine learning module, by extracting the feature of sample pictures, the system does modeling of insulators. Through modeling results, the system creates cascade classifier of insulators, which is used to located the position of insulator in the picture. In anomaly recognition module, the system gets angle of deflection by Gabor filter. Furtherly, the system determines whether there is abnormal insulator by the means of gray level statistic. In manual accurate detecting module, the paper describes the process of detecting abnormal insulators manually which is designed by the method of sequence diagram. In multi-core parallel detecting module, this paper introduces the application of parallel technology in the system based on OpenMP multi-core programming framework. 

Finally, this paper describes the development and testing environment. Besides, this paper analyzes the effect of each functional module of the system on the basis of different resolution images, different CPU cores and different application scenarios. Experiments show that the system can run efficiently on different CPU core computers. The system can detect anomaly of insulators automatically with higher processing speed and higher detection efficiency. The accurate detecting system is confirmed to has good human-computer interaction.

[1] 杨红梅. 基于图像处理的接触网支持及悬挂装置不良状态检测[D]. 西南交通大学, 2013
[2] 刘宇红. 绝缘子在运行中的常见故障原因及预防措施[J]. 内蒙古煤炭经济, 2016(1):114-116.
[3] 关志成. 绝缘子及输变电设备外绝缘[M]. 清华大学出版社, 2006.
[4] 高静辉. 输电线路绝缘子故障分析与检测方法综述[J]. 山东工业技术, 2016(15):165-166.
[5] 赵必武. 一种基于多传感器的接触网动态检测装置[C]// 2013 年中国铁路电气化技术装备交流大会暨2013 铁路电气化技术装备交流展示会. 2013.
[6] 高晓蓉, 王黎. 弓网故障动态检测装置的原理及应用[J]. 机车电传动, 1999(1):33-35.
[7] 岳路路. 基于机器学习的真菌孢子显微图像的特征提取与识别[D]. 西南大学, 2015.
[8] Freund Y, Schapire R E. A decision-theoretic generalization of on-line learning and an application to boosting[C]// European Conference on Computational Learning Theory. Springer-Verlag, 1995:119-139.
[9] Freund Y. Experiments with a new boosting algorithm[C]// Thirteenth International Conference on Machine Learning. 1996:148--156.
[10] 曹莹, 苗启广, 刘家辰,等. AdaBoost算法研究进展与展望[J]. 自动化学报, 2013, 39(6):745-758.
[11] 廖红文, 周德龙. AdaBoost及其改进算法综述[J]. 计算机系统应用, 2012, 21(5):240-244.
[12] 秦海兵. 基于多核平台的程序并行优化研究[D]. 长安大学, 2012.
[13] 冉晓龙. 基于多核多线程的混合并行编程技术研究[D]. 中原工学院, 2015.
[14] David R. Buten. Programming with POSIX Threads[M]. China Electric Power Press,2003
[15] 丁怡心,廖勇毅. 高斯模糊算法优化及实现[J].现代计算机, 2010(8):76-77.
[16] 张斌, 贺赛先. 基于Canny算子的边缘提取改善方法[J]. 红外技术, 2006, 28(3):165-169.
[17] Canny J.A Computational Approach to Edge Detection [J]. IEEE Transactions on Pattern Analysis and Machine Intelligence. 1986, 8(6)
[18] 李贞培, 李平, 郭新宇,等. 三种基于GDI+的图像灰度化实现方法[J]. 计算机技术与发展, 2009, 19(7):73-75.
[19] Duda R O, Hart P E. Use of the Hough transformation to detect lines and curves in pictures[J]. Ipsj Magazine, 1972, 15(1):11-15.
[20] 陈仁杰, 刘利刚, 董光昌. 图像主特征直线的检测算法[J]. 中国图象图形学报, 2010, 15(3):403-408.
[21] 陈盖凯. 基于Hough变换的直线检测[J]. 西安航空学院学报, 2007, 25(3):34-36.
[22] 吴铁洲, 熊才权. 直方图匹配图像增强技术的算法研究与实现[J]. 湖北工业大学学报, 2005, 20(2):59-61.
[23] Stephens M. A Combined Comer and Edge Detector[J]. 1988.
[24] Lowe D G. Distinctive Image Features from Scale-Invariant Keypoints[J]. International Journal of Computer Vision, 2004, 60(2):91-110.
[25] 史海山, 吕厚余, 仲元红,等. 基于遗传神经网络的火灾图像识别及应用[J]. 计算机科学, 2006, 33(11):233-236.
[26] 李志强, 李永斌. 车牌识别技术的发展及研究现状[J]. 科技信息, 2012(5):110-110.
[27] 尹义龙, 宁新宝, 张晓梅. 自动指纹识别技术的发展与应用[J]. 南京大学学报自然科学, 2002, 38(1):29-35.
[28] 邓耀华. 白细胞显微图像识别技术研究[D]. 广东工业大学, 2004.
[29] 杨凌云. 基于多核技术的并行图像检索系统的研究[D]. 北京化工大学, 2009.
[30] Viola P, Jones M J. Robust Real-Time Object Detection[C]// International Workshop on Statistical and Computational Theories of Vision -- Modeling, Learning, Computing, and Sampling. 2001:87.
[31] 曹健. 人脸检测和识别系统的设计与应用[D]. 南京信息工程大学, 2013.
[32] 林鹏. 基于Adaboost算法的人脸检测研究及实现[D]. 西安理工大学, 2007.
[33] 李彦冬, 雷航. 多核操作系统发展综述[J]. 计算机应用研究, 2011, 28(9):3215-3219.
[34] 汪松, 王俊平, 万国挺,等. 基于SIFT算法的图像匹配方法[J]. 吉林大学学报(工), 2013, v.43(S1):279-282.
[35] 韩本慧, 李莹莹, 付萌. 图像降噪算法研究[J]. 城市建设理论研究:电子版. 2015(8).