中文摘要：

针对训练样本在训练过程中的不同作用和支持向量机在推广到多类问题时存在不可分区域，可构造两类模糊支持向量机，其分别克服了过学习问题和减少了不可分区域．构造两类问题和多类问题综合的改进模糊支持向量机并用于手写数字识别，训练时，其利用数据与其类中心的相对距离定义隶属函数，测试时，利用S．Abe定义的隶属函数判别其类别．实验结果表明，该学习机具有比传统支持向量机和模糊支持向量机更高的精度．

英文摘要：

According to the different pole of different training sample in training process and the unclassifiable regions that exist during extension process,two kinds of fuzzy support vector machines （FSVMs）,which can overcome overfitting problem and reduce the unclassifiable regions,are constructed. The integration of two-class problem and multi-class problem FSVMs is proposed and used to the recognition of hand-written digit. During training process,the membership functions of two-class FSVMs are defined by using the distance between the training data points and their class center ,during decision process ,the membership functions of multi-class FSVMs defined by S. Abe are used to determine the class of testing data. The experiment results show that the improved learning machines can achieve the higher precision compared with the traditional SVMs and FSVMs