Abstract: Be aimed at the automatic choosing function of support vector machine on feature selection，we submit an improved minimal kernel classifiers.So fewer feature is used in the tests of unknown sample and the calculate amounts reduce in the computational time.Experiment results show that the algorithm is valid in suppressing the irrelevant features，which demonstrates its effectiveness on generalization ability.

摘要： 针对支持向量机在特征选择方面具有自动选择的功能，提出了一种改进的最少核分类器。在样本测试中使用更少的特征维数，减少识别过程计算量。数值试验表明，改进过的分类器能有效压缩无用的特征属性，具有较强的泛化能力。