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.