一种用于大规模模式识别问题的神经网络算法  

A Neural Network Algorithm for Large Scale Pattern Recognition Problems

作　　者：吴鸣锐[1] 张钹[2] 

机构地区：[1]清华大学计算机科学与技术系,北京100084 [2]清华大学智能技术与系统国家重点实验室,北京100084

出　　处：《软件学报》2001年第6期851-855,共5页Journal of Software

基　　金：国家自然科学基金资助项目 (6 982 30 0 1);国家重点基础研究发展规划 973资助项目 (G19980 30 5 0 9);高等学校博士学科点专项科研&

摘　　要：许多实际的模式识别问题如对手写体汉字的识别 ,都属于大规模的模式识别问题 .目前 ,传统的神经网络算法对这类问题尚无有效的解决办法 .在球邻域模型的基础上提出一种可用于大规模模式识别问题的神经网络训练算法 ,试图加强神经网络解决大规模问题的能力 ,并用手写体汉字识别问题检验其效果 .Many practical pattern recognition problems, such as recognition of handwritten Chinese characters belong to the pattern recognition problems of large scale. Now conventional ANN (artificial neural network) algorithms cannot solve this set of problems efficiently. In this paper, a neural network algorithm based on the sphere neighborhood model is introduced, aiming at enhancing the neural network's ability to solve the pattern recognition problems of large scale. The performance of the algorithm is tested with the handwritten Chinese character recognition problem. Experimental results show that the proposed algorithm is competent and has well prospects to this set of problems.

关 键 词：神经网络 模式识别 字符识别 球邻域模型 汉字识别 算法 

分 类 号：TP391.43[自动化与计算机技术—计算机应用技术] TP183[自动化与计算机技术—计算机科学与技术]