传统的非参数树分类器设计都采用先划分再编辑(即按—区域内各类训练模式的多少把这区域划归多数类的决策域)的方法。这种设计路线限制了识别率的提高。本文给出的先编辑训练样本再划分空间的设计方法,可以使树分类器达到依概率渐近圾优的识别率.相应的计算机仿真试验表明,这种树分类器在模式分布复杂和训练样本容量较小的情况下仍能达到较高的识别率;而且它的时间效率比其他非参数分类器,如近邻式分类器高出100至1000倍。 The traditional non-parametric tree classifier design adopts the method of dividing and re-editing (that is, how many different types of training modes in the region classify this region into the majority class decision region). This design approach limits the recognition rate. In this paper, firstly, the method of designing the training samples and then dividing the space can make the tree classifier achieve the recognition rate asymptotic probability of juxtaposition.Computer simulations show that the tree classifier has the following advantages in the complicated pattern distribution and training samples It can still achieve higher recognition rate with smaller capacity and its time efficiency is 100 to 1000 times higher than that of other non-parametric classifiers, such as the nearest neighbor classifier.