论文部分内容阅读
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。
The current SVM multi-class classification method has been widely used in the field of pattern recognition. In order to meet the real-time requirement of recognition system for multi-classification tasks, real-time implementation of SVM multi-class classification algorithm based on binary tree is studied. Combined with specific hardware implementation, a parallel method of decomposing a binary tree into several binary trees is proposed. A parallel real-time implementation scheme based on TMS320C6711 DSPs is given, that is, a binary tree SVM is completed with each DSP chip. And the interface circuit design and program flow are described. The implementation method has good fastness and scalability, and can be applied to systems requiring high real-time requirements and large numbers of categories.