【摘要】 针对原始K均值聚类算法的局限性,设计了一种改进的K均值聚类算法,结合不同睡眠阶段的特性,实现睡眠阶段的自动分期。首先,针对原始K均值选取初始聚类中心的随机性,基于密度和距离两项指标来选取初始聚类中心,使得初始聚类中心的选择更加合理,从而提高算法的稳定性;其次,选用高斯核函数作为聚类中心更新时的权值,减少离群点对中心的影响;然后,根据不同睡眠状态的特征,设计了分步聚类处理方式;最后,定义了距离修正系数对K均值聚类的结果加以修正,使其聚类结果更符合实际睡眠状态变化规律。将改进算法分别在来自不同数据集的睡眠数据上进行了测试,比原始K均值聚类有显著提升,且更贴近人工判读,能够为睡眠状态分析提供可行的辅助判读方式。更多还原

【Abstract】 Due to the limitations of the original K-means clustering algorithm,an improved K-means clustering algorithm was developed for automatic sleep staging by considering the characteristics of sleep stages. Unlike the randomly selected initial clustering centers,two indexes based on density and distance were designed to make the determination of initial cluster centers more reasonable and improve the stability of the algorithm. The Gaussian kernel function was adopted as the weight to update the cluster centers,which can reduce the influence of the outliers. Based on the characteristics of sleep states,a stepped clustering processing was implemented. The distance correction coefficients were defined according to the actual sleep states regulation to modify the clustering results. The improved K-means clustering algorithm was tested on the sleep data from different datasets respectively. Comparing with the original K-means clustering algorithm,the classification performance was improved. The obtained results were more close to the visual scoring by clinicians. The presented method can provide a feasible assistant way for sleep state analysis.更多还原