用于字符识别的椭圆基函数概率神经网络

ELLIPTIC BASIS FUNCTION PROBABILISTIC NEURAL NETEORK FOR CHARACTER RECOGNITION

提出了一种适合在有噪声条件下进行字符识别的椭圆基函数概率神经网络ＥＢＰＮＮ（ＥｌｉｐｔｉｃＢａｓｉｓＦｕｎｃｔｉｏｎＰｒｏｂａｂｌｉｌｓｔｉｃＮｅｕｒａｌＮｅｔｗｏｒｋ）模型，该模型选用由字符统计特征提取的具有选择注意特性的参数作为概率神经网络的椭圆基函数系数．ＥＢＰＮＮ模型在货车编号自动识别系统中获得了良好的应用，整个系统的字符识别率达到９６％以上，编号识别率达到９０％以上，实验结果表明该模型的识别性能较常用的其它神经网络要好，且特别适用于有噪声的情况．

An Elliptic Basis Function Probabilistic Neural Network (EBPNN)model for character recognition with noise was proposed, which uses selective attentional parameters extracted from statistic features of characters as elliptic basis function parameters. EBPNN is not only used for binary pattern recognition, but also for continuous pattern recognition. The experiments show that the recognition rate of EBPNN is better than that of other neural networks, especially under the noise circumstances.