提出一种基于半模糊矢量量化 (SFVQ)技术的改进径向基函数神经网 (IRBFNN)分类器 ,并且用于无约束手写体数字的识别 .作者在模糊聚类和矢量量化的基础上利用半模糊的思想提出了半模糊矢量量化算法 ,并在其中加入了有监督的控制 ,从而使系统在聚类过程中可以确定比较合适的类别数并使聚类结果能更好地反映训练集的概率分布 .以半模糊矢量量化作为预处理的改进 RBF网 ,应用了多尺度补偿等办法 ,能够充分利用训练样本集的信息 ,在整体和局部上都获得满意的训练效果 .作者对 N IST字库和实际采集的手写体数字样本的实验 ,表明此算法是令人满意的 This paper proposes an improved radial basis function neural network (IRBFNN) classifier based on semi-fuzzy vector quantization (SFVQ) technology, and is used to identify unconstrained handwritten digits. Based on fuzzy clustering and vector quantization, The semi-fuzzy vector quantization algorithm is put forward and the supervised control is added into it, so that the system can determine the appropriate number of categories in the clustering process and make the clustering result better reflect the probability distribution of the training set Using semi-fuzzy vector quantization as preprocessing improved RBF neural network, multi-scale compensation and other methods are applied to make full use of the information of the training sample set to obtain the satisfactory training effect on the whole and in part.At the same time, The experiment of collected handwritten digital samples shows that the algorithm is satisfactory