基于自增长型多级自组织映射网络的模式识别

Pattern Recognition Based on Self-Growing Multilevel Self-Organizing Map

以自组织映射网络为主要研完对象,描述了自组织映射网络的基本模型。在传统自组织映射网络的基础上,提出了基于自增长型多级自组织映射网络的模式识别方法,能够解决传统自组织映射网的静态结构带来的诸多问题,如在进行训练前必须预先确定网络的模型和神经元数目及其排列方式,若一次分类不准确将严重影响分析结果,等等。而且这种多组结构,还能将输入数据中存在的分级信息直观地表示出来,对于高维数据的分析尤其有利,因此自增长型多级自组织映射网络对大规模模式识别的研完一定会产生极大的促进作用。

Focusing on the self-organizing map, this paper presents a review on the basic model and puts forward a pattern recognition method that is based on self-Growing multilevel self-organizing map on the foundation of traditional self-organizing map. It can overcome many limitations, which are related to the static architecture of traditional model- For example, traditional model uses a fixed network architecture in terms of number and arrangement of neural processing elements, which has to be defined prior to training, also, if it has error in the classification for the first time, its results may not be corrected, etc. What is more, the new model can intuitively represent the hierarchical relations of the data, in particular, it can benefit the high dimension data analysis greatly. So, self-growing multilevel self-organizing map can promote the research of large scale pattern recognition greatly.