广义高斯核模型在心电信号稀疏分解中的应用

Generalized Gaussian Kernel Model and Its Application on the Decomposition of Electrocardiogram Signal

稀疏分解得到的心电信号表示结果对心电信号的压缩、特征提取等非常有利。根据心电信号的波形特点,提出了新的基于广义高斯函数的核模型。新的核模型通过正交最小二乘算法进行逐步回归建模,每一个核函数的中心,尺度及形状参数由重复加权提升搜索算法优化得到。为了避免过拟合,运用交叉验证的方法进行迭代终止时的阈值选择。实验数据来自MIT-BIH心电数据库,结果表明,对于心电信号,新的核模型具有稀疏性好,泛化能力高等优点。

Sparse decomposition has been widely used in the fields related to electrocardiograms（ECG）,such as compression and feature selection.According to the characteristic of ECG signal,this paper proposes a novel kernel model with generalized Gaussian kernel.Orthogonal least squares algorithm is utilized to construct the model term by term.At each regressor stage,the parameters in each term are tuned by repeated weighted boosting search.A cross-validation method is used to determine the termination threshold to avoid the possible overfitting problem.The numerical results show that the new model holds better sparsity and generality than the traditional ones.