正规化模糊神经网络及在手写体汉字识别中的应用  

Normal Fuzzy neural network and its application in script Chinese characters identification

作　　者：李盼池[1] 许少华[1] 

机构地区：[1]大庆石油学院计算机科学与工程学院,黑龙江大庆163318

出　　处：《计算机工程与设计》2005年第3期816-817,832,共3页Computer Engineering and Design

摘　　要：为改善手写体汉字识别的性能,提出了一种基于正规化模糊神经网络的识别方法。针对网络结构的优化问题给出了网络模型的规则层节点的选取方法和相应的反传播学习规则。该算法能够充分利用专家制订的“if—then”规则,完善网络的推理结构,提高网络的识别能力,减少噪声因素的影响。实验表明此方法对手写体汉字识别问题具有良好的适应性和实用性。该方法指出了一条进一步提高手写体汉字系统性能的新途径。In order to improve capability of script Chinese characters identification, a normal fuzzy neural network (NFNN) with five layers is proposed. Focusing on the structure optimization of network, a new node selection method and corresponding back propagation learning algorithm are presented this algorithm can sufficiently utilize fuzzy 'if-then' rules mapped out by expert, and it also can perfect reasoning structure of network, improve identification capability of the network and reduce influence of disadvantage factor. The ex- perimentindicates that this algorithm has a better adaptability and practicability forscriptChinese characters identification. This algorithm presents a novel approach of script Chinese characters identification capability improvement.

关 键 词：手写体汉字识别 模糊神经网络 学习规则 系统性能 算法 正规化 优化问题 传播学 对手 专家 

分 类 号：TP391[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]