基于DSP并行结构的二叉树SVM多分类器

目前SVM多类分类方法在模式识别领域得到了广泛使用。为了满足识别系统对完成多分类任务的实时性要求,研究了基于二叉树的SVM多类分类算法的实时实现。结合具体的硬件实现,提出了将一棵二叉树分解为几棵二叉树的并行方法。给出了基于TMS320C6711DSPs的并行实时实现方案,即用每一个DSP芯片完成一棵二叉树SVM的运算任务。并对接口电路设计和程序流程进行了阐述。该实现方法具有较好的快速性和可扩展性,可以应用于实时性要求较高和分类数目多的系统。

Binary Tree Multi - class SVM Classifier of Parallel Structure Based on DSP

The multi-class SVM has recently been extensively applied in the field of pattern recognition. Aiming at to meet the demand of multi-class mission in real time, this paper studies the real-time implementation of multi-class SVM based on binary tree. Combining with the concrete hardware implementation, a parallel method of dividing one binary tree into several binary trees is put forward. The scheme of implementing the multi-class classifier on TMS320C6711DSPs with parallel structure is put forth, where each DSP chip accomplishes one binary tree SVM computing task. The design of circuit and the program process are also expatiated. The result of simulation proves that this method has a better speedy capability and expansibility and could be used in lots of systems, especially the ones requiring high performance in real time with large number of classifiers.