基于RSOM树模型的机器学习原理与算法研究  

Theory and Algorithm of Machine Learning Based on RSOM Tree Model

作　　者：夏胜平[1] 张乐锋[1] 虞华[1] 张静[1] 胡卫东[1] 郁文贤[1] 

机构地区：[1]国防科学技术大学电子科学与工程学院ATR重点实验室,湖南长沙410073

出　　处：《电子学报》2005年第5期939-944,共6页Acta Electronica Sinica

基　　金：国家863计划(No.2003AA134030);国防科技预研项目(No.Y018-6/Q200210)

摘　　要：　机器学习和识别可归结于一个高速、有效地搜索非常大的样本空间问题,以实现对训练和识别样本的最佳拟合.对于复杂背景的模式样本集,同类型样本的独立同分布(i.i.d)特性通常难以保证,统计理论无法有效应用.本文将层次化思想和自组织映射(SOM)神经网络相结合,采用递归实现技术实现了一种高效、高容量,能够自适应增长的模式分类树(RSOM树)生长方法,用于模式识别和机器学习的基本建模.通过对大量公用数据集的测试以及在实际的雷达目标识别系统中应用,方法有效性得到了证明.One useful perspective on machine learning and target recognizing is that they involve effectively and rapidly searching some very large sample space to determine one that best fits the observed data and trained samples. If the independent identical distribution (i.i.d) property of pattern samples of complexity background is not satisfied, statistic theory is of no effect. In this paper, a method on producing RSOM pattern recognition tree is proposed, in which the hierarchy model, SOM neural net algorithm and recursive technique are utilized. This is a basic model of pattern recognition in machine learning. The validity of this method is proven by testing lots of common data sets and practical application in radar target recognition system.

关 键 词：模式识别 分类树 神经网络 SOM RSOM 机器学习 

分 类 号：TP8[自动化与计算机技术—检测技术与自动化装置]