【作者】 李贵娟；

【导师】 赵奇；

【作者基本信息】 河北工程大学 ， 计算机应用技术， 2011， 硕士

【摘要】 心血管疾病是当今危害人类健康的主要疾病之一,心电图检查是临床上诊断心血管疾病的重要方法。心电图准确的自动分类对于心血管疾病的诊断起着关键作用。聚类分析是非监督模式识别的一个重要分支,它是用数学的方法研究和处理给定对象分类。模糊聚类建立在样本对于类别的不确定性描述的基础上,更能客观的反映现实世界,从而成为聚类分析研究的主流,并在许多领域得到了广泛的应用。目前,已提出了许多模糊聚类算法,其中最常用的是基于目标函数的模糊c-均值聚类算法(FCM)。针对此算法中存在的需要聚类先验知识的问题,采用SOM神经网络算法作为FCM算法的先导级,先将样本经过SOM神经网络的训练,得到聚类类别数,但此方法得到的类别数与实际结果存在较大偏差。因此提出了一种改进方法,即将SOM神经网络、优化的系统聚类法和FCM算法相结合的聚类方法。首先对系统聚类法进行优化,然后使用优化后的系统聚类法分析SOM神经网络初始分类的结果,最终得到更合理的聚类类别数和聚类中心,将此聚类数和聚类中心用于FCM算法的输入进行进一步聚类,从而得到精确的聚类信息。最后,采用MIT/BIH心电数据库中的数据来仿真,结果说明此种方法具有很好的聚类效果。更多还原

【Abstract】 Cardiovascular disease is one of major disease which endangers human’s health. The analysis of ECG(Electrocardiogram) is an important means for diagnosing cardiovascular disease in clinic. The accurate automatic classification of ECG plays a key role for the diagnosis of cardiovascular disease.The cluster analysis is an important branch of non-supervision pattern recognition, which uses the methods of mathematics to research and deal with given classification of objects.Fuzzy clustering based on the uncertain description for sample to category, which can more objectively reflect the real world, thus becoming the mainstream of cluster analysis, and has been widely used in many areas.At present, many algorithms for fuzzy clustering are proposed, in which the most commonly used is the fuzzy c-means clustering algorithm based on the objective function.For the problem of the algorithm which requires clustering prior knowledge, we take the SOM neural network algorithm as the forerunner of FCM algorithm. Using the SOM neural network to train samples, and get the number of clustering categories. But it has a big deviation in the actual results and the results for using this method. Therefore, this paper proposes an improved method, which is a combination clustering method of the SOM neural network, the optimizing system algorithm and the FCM algorithm. First, optimizing the system clustering method, then using the optimized system clustering method to analyze the initial classification results of SOM neural network, ultimately get more reasonable number of cluster categories and cluster centers, through using the number of cluster categories and cluster centers as the input of FCM clustering algorithm to further cluster, so that get accurate clustering information.Finally, take the data in the MIT / BIH ECG database to simulate, and show that this method has good clustering effect through the results.更多还原