一种基于球邻域模型的神经网络算法  

A Neural Network Algorithm Based on Sphere Neighborhood Model

作　　者：顾磊[1] 宫宁生[1] 朱梧槚[2] 

机构地区：[1]南京工业大学信息科学与工程学院,南京210009 [2]南京航空航天大学计算机科学研究所,南京210016

出　　处：《模式识别与人工智能》2005年第6期747-751,共5页Pattern Recognition and Artificial Intelligence

基　　金：国家重点基础研究发展规划(973)基金(No.G199903270)

摘　　要：手写体汉字识别问题属于一种大规模的模式识别问题。本文基于球邻域模型的几何意义解释,即将神经网络的训练转化为几何的点集覆盖问题,通过对神经网络分界面的分析提出了一种改进的前馈神经网络训练算法,并且引入神经网络集成的思想,用以解决手写体汉字的识别问题,实验结果表明该算法可以用来解决大规模的模式识别问题且具有较好的效果。Recognition of handwritten Chinese characters belongs to the pattern recognition problems in large scale. This paper firstly presents a geometrical representation of sphere neighborhood model, by which the training problem of neural network may be transformed into the geometrical covering problem of a point set. After the neural network separate boundary faces are analyzed, an improved training approach to feedforward neural network is introduced and neural network ensemble is also applied. The performance of the approach is tested with the handwritten Chinese characters recognition problem . Laboratorial results not only are satisfactory but also show that the proposed approach is effective in the pattern recognition problems in large scale.

关 键 词：神经网络 手写体汉字识别 训练算法 球邻域模型 神经网络集成 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]