基于SGNN的数字字符识别

Digit Recognition Using Self-Generating Neural Network

自生成神经网络SGNN具有自主学习能力强和计算速度快的特点,可用于识别带噪声的数字字符。首先提取数字字符的特征矢量,然后将特征矢量输入SGNN中对SGNN进行训练建立分类器,通过比较未知样本特征矢量和分类器根节点权值矢量的距离远近从而得到识别结果。实验表明这种方法有较高的识别正确率,其性能优于BP神经网络。

Using SGNN(Self-Generating Neural Network)to solve the problem of digit recognition is presented in this paper. SGNN's learning is an unsupervised process by which we could train a SGNT(Self-Generating Neural Tree)with a kind of digit's (such as 0,1,etc.)feature vector rapidly. Because each SGNT represent a number(0,1,…,9), we could find out the SGNT which has the minimum distance among 10 SGNTs —the distance is calculated from testing sample and SGNT's root—then the testing sample could be decided what number it is . The simulation proves that this method has a good performance in digit recognition.