Abstract: Granular computing is a method for simulating human cognition to solve problems from different granularities and levels, and has become an important research direction in intelligent information processing. As one of granular computing methods, Cloud model is a qualitative and quantitative transformation cognition model based on probability theory. The present cloud model algorithms cannot resolve how to extract multiple concepts with different granularities from a data set on problem domain. Gaussian Mixture Model（GMM） is an important mathematical model, which can transfer an arbitrary probability distribution to sum of some Gaussian distributions. Based on this, this paper proposes a measurement of the confusion degree of concepts based on the atomized feature, and designs the Gaussian Cloud Transformation algorithm（GCT）. GCT transfers a data set on problem domain to some concepts on different granularities automatically, and pan concept tree  is also formed in this process. Applications to data clustering and image concept extraction are presented to validate the effectiveness of the proposed method.

Key words: granular computing, Cloud model, Gaussian Mixture Model（GMM）, concepts extraction

摘要： 粒计算是研究和模拟人类认知从多粒度、多层次解决问题的方法，近年来成为智能信息处理中一个热点方向。云模型是一个基于概率理论研究定性定量转换认知模型的粒计算方法，通过正向和逆向云算法实现一组数据样本和一个基本概念之间的转换，但是目前的算法不能在整个问题域中解决多粒度、多概念的生成问题。概率统计中的高斯混合模型可以将任何一个频率分布函数转换成多个高斯分布的叠加，在此基础上，创新地提出用云模型中数字特征构建概念含混度作为概念外延共识程度的衡量，设计并实现了高斯云变换算法，将问题域中的数据分布自动转换为多粒度的不同概念，构建出人类概念认知中的泛概念树。通过在数据概念聚类和图像分割中的应用，验证了方法的有效性。

关键词: 粒计算, 云模型, 高斯混合模型, 概念抽取