上海交通大学学报

上海交通大学学报（自然版）

• 自动化技术、计算机技术 • 上一篇    下一篇

基于动态神经网络支持向量机的FPGA实现

刘涵，尹嵩，刘丁   

  1. (西安理工大学 自动化与信息工程学院， 西安 710048)
  • 收稿日期:2009-09-09 修回日期:1900-01-01 出版日期:2010-07-28 发布日期:2010-07-28

FPGA Implementation of Dynamic Neural Network for Support Vector Machines

LIU Han,YIN Song,LIU Ding   

  1. (School of Automation and Information Engineering, Xi’an University of Technology, Xi’an 710048, China)
  • Received:2009-09-09 Revised:1900-01-01 Online:2010-07-28 Published:2010-07-28

摘要： 研究了一种基于动态神经网络支持向量机(SVM)的FPGA硬件实现方法.提出了基于动态神经网络的最小二乘支持向量机(LSSVM)神经网络结构，完成了VHDL语言描述的基于动态神经网络的LSSVM结构设计，并在XILINX SPANT3E系列FPGA中完成了LSSVM的分类与回归实验.结果表明，该硬件实现方法很好地完成了SVM的分类与回归功能，与现有的软件仿真和模拟器件实现相比，该方法具有更快的收敛速度和更高的灵活性.

关键词: 支持向量机, 最小二乘支持向量机, 动态神经网络, 稳定性

Abstract: A new FPGA hardware implementation approach of dynamic neural network for support vector machines was provided and researched.The structure of dynamic neural network for least square support vector machines (LSSVM) was proposed. The architecture design of dynamic neural network for LSSVM based on VHDL language was also performed. The experiments of classification and regression for LSSVM were achieved on XILINX SPANT3E series FPGA. The experimental results show that it is effective to complete the LSSVM classification and regression based on presented method. Compared with the existing methods based on software implementation or analog device implementation, this approach has better convergence rate and better flexibility.