支持向量机中Mercer核函数的构造研究

Research on Construction of Mercer Kernel-Function in Support Vector Machine

核函数的构造是支持向量机中的一个重要的问题,它直接影响模型的推广能力。在核匹配追踪学习机的基础上,提出子波核函数和多尺度核函数,利用多分辨分析可逼近平方可积空间L2(R)的性质,弥补传统核函数在逼近性能方面的不足,并通过实例证明其有效性。

The construction of kernel-function play an important role in support vector machine. It has a great influence on the model application. Propose the wavelet Mercer kernel and MRA Mercer kernel based on kernel matching tracing learning machine, with the feature of MRA can approach Square L2（R） to cover the shortage of traditional Mercer-function and the two Mercer kernels are verified by experiments.