基于吴方法的非线性分类器设计  

Design of Nonlinear Classifier Based on Wu's Method

作　　者：聂一鸣[1] 安向京[1] 贺汉根[1] 常文森[1] 

机构地区：[1]国防科技大学自动化研究所,湖南长沙410073

出　　处：《计算机仿真》2007年第8期137-140,149,共5页Computer Simulation

摘　　要：在模式识别领域工程实际中占大多数的是非线性问题,传统的分类器设计方法是将非线性问题转化为线性问题解决,但是这样不能很好的解决日益复杂化的非线性问题。文中提出的新方法为解决非线性问题提供一个新的思路和手段。该方法立足于直接处理非线性问题的吴文俊方法,进行非线性分类器的设计。经过选择分类函数的复杂度和训练点,建立非线性方程组,使用吴方法解出这个非线性方程组,得到分类函数。使用该方法进行异或问题和鸢尾花数据的实例研究,并与神经网络方法和支持向量机方法进行比较分析。该方法是使得结构风险和经验风险总体最小化的方法,并且过程直接、简洁、运算量小。该方法尽管有不完善的部分,但是不失为有潜力发展的新方法。In Statistical Pattern Recognition the main problem is nonlinear. The traditional method of nonlinear classifier is to convert nonlinear problems into linear problems, but it＇ s not suitable for more and more complex nonlinear problems. A new nonlinear classifier design method is introduced, based on Wu＇s method which deals with nonlinear problems. This method provides a new means to solve the nonlinear problem. A system of nonlinear equations is created by choosing the degree of polynomials and limit points. The classifier function is formed to solve the system of nonlinear equations by means of Wu＇ s method. The XOR problem and IRIS data sets are investigated and compared with Neural Networks and Support Vector Machine. This method minimizes the structural risk and experience risk. It＇ s direct and simple. Though this method is not perfect, it is a developing method with hopes.

关 键 词：统计模式识别 吴方法 非线性 神经网络 支持向量机文献标识码:A 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]