【作者】 袁晓燕；

【导师】 魏世泽； 刘教民；

【作者基本信息】 河北工业大学 ， 计算机应用技术， 2004， 硕士

【摘要】 低压电器电弧的运动对电器的分断能力、电寿命以及判断电器是否发生故障都起着相当重要的作用。揭示电弧从燃弧到灭弧的动态变化过程，是提高低压电器产品可靠性的有效途径之一。 边缘检测是数字图像处理中一项非常重要的技术。对电弧图像进行边缘检测是研究电弧运动的前期工作，能否找到一种适当的且效果好的边缘检测算法，在对电弧运动的研究中至关重要。本课题主要就是针对低压电器的电弧图像进行边缘检测的算法研究。 本文介绍了与课题相关的各种理论、概念，如图像分割、边缘提取以及模式识别的理论和常用方法等，并对数字图像处理中常用的方法、函数等也做了介绍。 主要的研究工作包括以下几个方面： 对一些传统的边缘检测算子，如：Roberts算子、Sobel算子、拉普拉斯算子等等进行了原理分析，并将这些算子应用于电弧图像，测试其边缘检测的效果。但这些传统边缘检测算子都过于简单，其去除噪声以及确定边缘位置等性能不是很理想。本文也对高斯．拉普拉斯(LoG)算子进行了分析，但经过分析测试其边缘检测效果依然不是很令人满意。 John Canny提出了边缘检测的三条准则，并在此基础上提出了一个效果很好的算法：Canny算法。本文详细介绍分析了这三条准则和Canny算法，并对Canny算法进行了实现并测试其效果。 然后，本文对以上算法进行了详细的评估与对比。并在这些算法的基础上，基于电弧图像的特点，提出了一种改进的多阶段边缘检测算法，这个算法中应用到模式识别中的类别可分离判据的理论，并使用了Canny算法所中提出的双阈值法和非极大值抑制方法等等。 测试该算法在电弧图像上能够取得较好的边缘检测效果。更多还原

【Abstract】 The arc moving process of low voltage apparatus plays an important role for the disjunction ability of the contact, electrical life and its cause of invalidation. To improve its reliability, it is effective to investigate the dynamic change of the whole process from arcing to arc suppress.Edge detection is an important part of the digital image processing. Detecting the edge of the image of the electrical arc is the precondition for the research of the dynamic change of the arc. So it is especially necessary for this research to find an appropriate detection algorithm with a good detecting result. This subject choose arc image as the research object.This paper introduces various theories and methods relevant to this subject, such as image segmentation, edge detection, pattern recognition etc. And this paper also introduces some functions and methods often used in digital image processing.Some conventional detection operators are analysed in detail, such as Roberts operator, Sobel operator and Laplacian operator, and also apply this operators to the arc image to test their detection results. But these conventional detection operators have a common shortcoming: they are too simple. This made their performance of noise suppressing and edge orientation is not very good. We also analyse LoG operator in this way. Although it is more complicated, it also cannot get a satisfactory result.John Canny presented Canny criteria for edge detection, and also presented an algorithm based on these criteria, named Canny operator. In this paper, we analyse Canny criteria and Canny operator in detail.We compare and evaluate the algorithms mentioned above. Based on these algorithms, and consider the property of the arc image, we proposed a new multi-stage algorithm. The proposed algorithm involves the class separability criterion in pattern recognition, and double-threshold method and non-maximum suppression proposed in Canny operator. We apply this algorithm to the arc image, and get a better result.更多还原