【作者】 刘莉；

【导师】 程琼；

【作者基本信息】 湖北工业大学 ， 电力电子与电力传动， 2014， 硕士

【摘要】 本文主要研究利用图像分割技术来提高复合绝缘子憎水性等级判定的准确率。文章围绕复合绝缘子憎水性图像分割中的三个问题展开，有针对性地对三个问题提出解决方案，有效提高复合绝缘子憎水性图像分割的准确性，进而提高憎水性等级判定的准确率。在了解目前复合绝缘子憎水性状态研究的基本概况、在分析传统遗传算法的图像分割基础上，归纳总结了复合绝缘子憎水性图像分割中的问题：①一种算法无法适用于所有憎水性等级；②相邻水珠会出现过识别现象；③光照造成的图像水珠存在缺口问题。对问题形成的原因进行理论分析，主要原因在于复合绝缘子憎水性图像受光照的影响，加上水珠自身的透明性等因素，导致图像出现灰度分布不均，无法找到统一的分割阈值适用于所有等级的所有憎水性图像。因图像直方图充分反映了图像中各像素点的灰度值，本文将直方图峰值特征作为一个新的特征量，提出了两种基于图像直方图的分割算法。算法一，采用基于直方图峰值与遗传算法结合的图像分割算法，以解决传统遗传算法无法适用于所有憎水性等级的问题。算法二，将直方图引入区域生长算法中，运用图像直方图的灰度分布特性实现种子点的自动选取，即采用直方图的自动种子区域生长算法，能较好的解决多个相邻的单个水珠识别成一个较大水珠的问题。对于解决图像水珠存在缺口的问题，本文提出了算法三，基于prewitt边缘检测与图像融合的分割算法。通过对各种常用边缘检测算子的比较，得出prewitt边缘检测算子最适合提取光斑或光迹的边缘信息。采用的改进prewitt边缘检测算法，对存在缺口问题的憎水性图像，能得到范围准确、轮廓清晰的光斑或光迹边缘。再利用图像融合技术，将光斑或光迹的图像分割效果图与初识别中存在缺口问题的效果图进行合并。该方法能有效地分割憎水性图像水珠形状信息，较好地解决了因光照造成水珠缺口的问题。本研究提出的三种图像分割方法均根据实际问题提出，并结合用于复合绝缘子憎水性状态分析中，可满足不同情况下的憎水性图像分割要求。实验结果分析表明，本研究能分割出边缘连续、区域较完整的憎水性图像水珠，三种算法结合使用能使憎水性等级判定的准确率达到95%以上。更多还原

【Abstract】 The main purpose of this study is to improve the accuracy of judging the level ofhydrophobicity of composite insulators, by using image segmentation techniques.Through analyzing the three issues of hydrophobic image of composite insulators inthe image segmentation process, the paper puts forward three solutions. It caneffectively improve the accuracy of image segmentation and grade determination forhydrophobic image of composite insulators.Through understanding the basic situation of hydrophobicity of compositeinsulators, and through the analysis of the traditional genetic algorithm, This papersummarized three problems of composite insulator hydrophobic image in imagesegmentation. Namely: an algorithm is not suitable for all level of hydrophobic;Adjacent water may form a large droplets of water in the process of recognition; Waterdroplets in the hydrophobic image will appear gap caused by light. Through theoreticalanalysis, The main causes of the problem is that the hydrophobic image of compositeinsulator will be affected by light and the transparency of the water. So that the graydistribution of image will be uneven, and it will lead that the image segmentationalgorithm can not find a unified segmentation threshold to apply to all levels of allhydrophobic image. Considering the image histogram fully reflects the gray value ofeach pixel in the image, We put the histogram peak as a new feature and put forwardtwo kinds of image segmentation algorithm based on histogram. One is: we use animage segmentation algorithm based on histogram peak and genetic algorithm, to solvethe problem that the traditional genetic algorithm is not suitable for all level ofhydrophobic image. Another is: we propose a region growing algorithm which canautomatically select a seed point according to the histogram distribution of gray image,to solve the problem that adjacent water may form a large droplets of water.To solve that, water droplets images have gap because of light, this paper presentsa third algorithm based on prewitt edge detection and image fusion. By comparing allkinds of edge detection operators, we found that the prewitt edge detection operator issuitable for extracting the edge information of the image spot drops. The improvedprewitt edge detection algorithm can get a clear outline of the edge of the water dropletsspot, when dealing with a gap problem of hydrophobic image. At the same time, thealgorithm also uses image fusion technology, to combine the spot image segmentationresult and the original image segmentation result. This method can get a complete dropsshape of hydrophobic image, and it solved the gap problem of hydrophobic image.This study presents three image segmentation methods based on the actual problem,to meet the requirements of hydrophobic image segmentation in different situations. Theexperimental results show that the results of image segmentation meet the researchrequirements in this study. We can find that the water edge is continuous and the waterarea is complete in image segmentation results. The accuracy rate of determining thelevel of hydrophobic can reach more than95%, when combined with the threealgorithms.更多还原