摘要
将分支前馈神经网络(BFNN)运用于数字字符的模式识别问题中,其某些性能优于标准反向传播(BP)网络。BFNN的隐层神经元与输出神经元之间为分组对应关系,采用的学习算法与标准BP算法类似。BFNN可以根据样本的可分性构建最适宜的网络结构。在对大规模、分类复杂的样本进行识别时,性能优于标准BP网络。
Utilizing the branched feedforward neural network （BFNN） to deal with the problem of recognition of number characters is better than that of standard BP network in some performance. The BFNN＇s hidden layer neurons are divided into several groups, and each hidden layer connects with only one output neurons. The Learning algorithm used by BFNN is similar to BP＇s. BFNN can build the suitable network structure by itself based on the training set. For the problem of the recognition of the large scale training set, the learning and testing performance of BFNN is better than that of standard BP network.
出处
《上海电机学院学报》
2006年第5期54-57,共4页
Journal of Shanghai Dianji University
基金
上海市教委科研资助项目(沪教科01D02-1)
关键词
分支前馈网络(BFNN)
模式识别
标准反向传播网络
数字字符
branched feedforward neural network （BFNN）
pattern recognition
standard back propagation network
number character
