中文摘要：

为了提高支持向量机（SVM）多分类器的训练效率，将多球体思想引入有指导学习，对训练样本按类别分别进行一类支持向量机（1-SVM）训练得到多球体分类器．针对多球体的冗余区域，构造简化一对多分类器将各球内混叠样本与正常样本分离．以上两个分类器性能互补，可以加权组合为多球体一对多分类器．同时给出了组合分类器基于交叉验证的权重估计和参数调整．仿真实验表明，相对于一对多算法，该分类器训练时间较短且分类正确率较高；相对于一对一算法，该分类器决策速度较快，有助于解决大样本的多分类问题．

英文摘要：

In order to improve the efficiency of the multi-class classifiers based on support vector machine （SVM）, the multi-sphere method was introduced to supervised learning. By training one-class SVM （1- SVM） on the samples class by class, a classifier composed of multiple spheres was obtained. To remove the redundant region in the spheres, a compacted one-vs-rest classifier was used to separate the mixed samples. These two complementary classifiers can be combined into a weighted classifier of one-vs-rest and multi-spheres. The regularization method of the weight factor and other parameters was given based on cross validation. Simulation showed that the novel classifier has higher accuracy with less training time when compared with one-vs-rest classifier, and its decision rate is faster than that of one-vs-one classifier. Consequently, the novel classifier is helpful for solving multi-class problems on large scale systems.