【摘要】 模式识别中分类树方法可用于提高模式分类的速度。该文分析了分类树的树结构对分类树差错率与速度增益的影响,得到差错率与速度增益之间的关系,进而给出一个调整分类树的优化准则,在此基础上提出了一种新的分类树生成策略。实验证实,采用此新算法生成的分类树进行模式分类时,与不使用分类树相比,在分类树差错率为1%的情况下,所需计算量只有原来的13.7%;与直接使用ISODATA聚类法生成的分类树相比,在分类树差错率同为1%的情况下,新算法生成的分类树的速度增益是原分类树的1.6倍。更多还原

【Abstract】 Classification trees are often used in pattern recognition to speed up the classification. The influence of the classification tree structure on the classification error rate and speed gain was analyzed to relate the error rate and the speed gain. An optimizing criterion for adjusting the classification tree structure was used to develop a new algorithm for constructing overlapped classification trees. For a tree misclassification error rate of 1%, the computation cost was only 13.7% of that not using the classification tree. The speed gain using the new classification tree algorithm was 1.6 times that using the ISODATA clustering method.更多还原