摘要
本文研究了径向基函数网络（ＲＢＦＮ）的分类机理问题。在Ｒｕｃｋ工作的基础上，通过与传统的基于Ｐａｒｚｅｎ窗估计核分类器做类比，本文从模式分类机理入手，分析了ＲＢＦＮ使用正、负两类训练样本来估计判别函数的特点，指出它优于核分类器，并讨论了相应情况下ＲＢＦＮ输出层连接权、模式分类判决域的特点。最后用多类模式分类的结果对上述理论进行了验证。
The classification mechanism of a radial basis function network(RBFN)is investigated in this paper. From Ruck's conclusion,for pattern classification based on the method of Parzen window estimation,it is shown in this paper that the RBFN has better performance than the conventional kernel classifier by using two kind of,that is,positive and negative training samples.The role of the weights of a RBFN in the estimation of decision function is described heuristically for multiclass problem.An example of classification experiment is presented to illustrate the analysis in the paper.
出处
《通信学报》
EI
CSCD
北大核心
1996年第2期86-93,共8页
Journal on Communications
基金
电子工业部电科院预研基金
关键词
神经网络
径向基函数
模式分类
核分类器
neural network,radial basis function,pattern classification,kernel classifier,Bayes probability
