【作者】 邱存勇；

【导师】 肖建；

【作者基本信息】 西南交通大学 ， 电力电子与电力传动， 2013， 博士

【摘要】 模糊聚类算法因其类人的逻辑语言及易于实现的优点称为了聚类分析的最主流研究方法,且在图像分割、大规模数据分析、数据挖掘、模式识别等众多领域中得到广泛应用。模糊理论为模糊聚类分析提供了理论基础,随着模糊理论的发展,一型模糊集处理不确定性能力差的缺陷逐渐显现,因此拥有更强处理不确定性能力的二型模糊集成为了模糊理论研究的热点。在模糊控制等领域中,二型模糊集已经表现出优于型模糊集的处理能力。然而在模糊聚类分析中,二型模糊集的应用仍处于起步阶段。虽然目前二型模糊集在模糊聚类分析中已经有成功地尝试,并提出了较为有效的区间二型和广义二型模糊聚类算法,但是由于二型模糊集的运算复杂度较高,直接制约了二型模糊聚类算法的发展与应用。为提升二型聚类算法的运算效率以及处理大数据量样本的能力,通过对区间二型模糊聚类算法的深入研究,提出了结合降型特点的初始化聚类中心方法,并对降型运算的初值选择和运算过程进行优化,消除了传统方法的计算冗余。通过大量的实验结果表明,优化的区间二型模糊聚类算法在运算效率上较传统算法有约40%的提升。由于模糊聚类分析处理的数据类型多种多样,单一的聚类算法无法满足不同数据的聚类要求,因此二型模糊聚类算法的推广需要结合不同的研究背景展开。图像分割作为模糊聚类分析的一个重要应用方向,是目标识别、图像理解、计算机视觉研究中最基本、最重要的处理步骤和关键技术。首先,对于现有的图像分割模糊聚类算法进行归纳总结,并分析了改进方法在区间二型模糊聚类中应用的可能性。同时,考虑二型模糊聚类算法的有效性评估处于空白状态,将一型模糊聚类算法的有效性函数进行了扩展,提出了几种适用于区间二型模糊聚类算法的广义有效性函数。此外,结合图像分割中图像像素点相互关联的特点,提出了描述像素点邻域信息的空间隶属度函数。通过引入空间隶属度函数的区间二型模糊聚类算法进行图像分割,提高了算法对于噪声点以及边缘处像素点的划分精度。对人造图像及医学图像的分割结果验证了所提算法的有效性。作为接触网与受电弓的唯一连接部件,受电弓滑板肩负着保障机车运行动力的使命。由于铁路系统的不断提速,受电弓滑板的损耗情况日益严重。及时检测受电弓滑板的磨耗状态并更换过限滑板是铁路安全检测的重要任务。随着智能化铁路监控系统的发展,数字图像处理技术成为了受电弓滑板状态检测的重要手段。而图像分割作为图像目标提取与后期处理的重要步骤,其分割结果的好坏直接影响图像处理的最终结果。传统图像检测算法在检测现场图像时易受噪声、天气、光照等众多因素影响,导致检测到的滑板边缘线段不连续,从而影响检测的精确度。因此,考虑将改进的区间二型模糊聚类算法引入到受电弓滑板检测中,通过模糊聚类算法抑制噪声等各种环境因素的干扰,以提升滑板边缘的检测精度。同样地,算法也引入到接触网杆号的检测中,通过对杆号的图像识别,有效地定位故障发生的地点,为故障维修提供了便利。实际图像的处理结果验证了引入模糊聚类图像分割方法的有效性。更多还原

【Abstract】 Because of its humanoid logic language and ease of implementation, fuzzy clustering becomes the most popular method among all clustering analysis methods and has been widely used in image segmentation, large-scale data analysis, data mining, pattern recognition, etc. Fuzzy theory provides a theoretical foundation for fuzzy clustering analysis. With the development of fuzzy theory, the defect of handling uncertainties in type-1fuzzy set gradually emerged over recent years. Therefore the type-2fuzzy set with ability of handling uncertainties became immediate areas of research focus. Type-2fuzzy set has demonstrated superior to type-1fuzzy set in application in fuzzy control and other areas. However, the use of type-2fuzzy set in fuzzy clustering analysis is still in its infancy.There are some researches on combining type-2fuzzy set with fuzzy clustering and have proposed interval type-2and general type-2fuzzy clustering algorithms. But the computation complexity in type-2fuzzy set restricts the development and application of type-2fuzzy clustering algorithms. Through the in-depth study of interval type-2fuzzy clustering algorithm, an optimization method is proposed in this paper to enhance the computing efficiency and ability of managing large-scale data. An initialization method for cluster center is proposed and the calculation process of type-reduction has been optimized to elimate calculating redundancy. A mass of experimental results show that the optimized interval type-2fuzzy clustering algorithm has about40%improvement in operation efficiency compared with traditional interval type-2algorithm.Fuzzy clustering always treats diverse data types, and single clustering algorithm cannot meet the requirements of clustering different data. As a result, the application of type-2fuzzy clustering algorithms requires combination of different research background. As a main direction of application for fuzzy clustering, image segmentation is the most important process in object recognition, image understanding, computer vision, et al. A summary for existing fuzzy clustering algorithms in image segmentation is given in this paper, and the utilization of existing modified methods in interval type-2algorithm is discussed either. Consider the validity evaluation for interval type-2fuzzy clustering is still blank, the validity functions for type-1fuzzy clustering are extended to a generalized version to evaluate interval type-2fuzzy clustering algorithms. Besides, by taking into account the relevance among pixels in images, the spatial membership function is proposed to represent spatial neighborhood information. The utilization of spatial membership function in interval type-2fuzzy clustering algorithm gives the algorithm the ability of better segmenting pixels with noise and pixels along the edges. The experiments on synthetic and medical images verified the effectiveness of the modified algorithm.As the only connection component between catenary and pantograph, pantograph slide plays an important role in providing impetus for locomotives. With the speed-up in railway operation, the abrasion of pantograph slide is getting worse. Thus it’s a vital task in railway safety inspection to detect the state of pantograph slide. With the development of intelligent railway monitoring system, digital image processing has become an important method to detect the state of pantograph slide. As a critical process in image object extraction and post-processing, the image segmentation results will directly affect the detection results. The traditional image detection methods are susceptible to noise, weather, lighting and other factors, which make the detected edges of pantograph slide discontinuous. To enhance the ability of suppressing the interference of noise and other environmental factors as well as to improve the detection accuracy, the modified interval type-2fuzzy clustering algorithm is used to segment pantograph images. Besides, the algorithm is utilized in the identification of catenary pole number to locate the failt point. The detection results validate the advantages of using interval type-2fuzzy clustering image segmentation process.更多还原