关键词: 支持向量机, Mercer核, 特征空间, 二叉树, 多类分类

Abstract: This paper proposes a new SVM multi-classification method utilizing the kernel theory.To get better separability，the input space is mapped to a high-dimensional feature space（Hilbert） applying Mercer kernel function.With a suitable choice of the kernel，the data can become separable in feature space despite being non-separable in the original input space.Then the hypersphere class least cover is used to be the rules of constructing binary tree.Classification experiments prove that the improved algorithm has better classifying performance than the other methods，and resolve the problem of nonlinear separability of classification in input space effectively.

Key words: Support Vector Machine（SVM）, Mercer kernel, feature space, binary tree, multi-classification