论文部分内容阅读
基于核学习的非线性映射能力,提出一种小波核广义方差的核独立成分分析算法WKGV-KICA.小波核函数具有近似正交,适用于信号局部分析的优点.与互信息相联系,将核广义方差作为对比函数对统计独立性进行衡量,可以获得理想的数学特性.将该算法应用于宽范围的盲源分离问题的实例中,并与现有算法进了比较.实验结果表明,WKGV-KICA算法在同等条件下的分离精度更高,而且性能更好.
Based on the nonlinear mapping ability of kernel learning, a kernel independent component analysis algorithm WKGV-KICA for wavelet kernel generalized variance is proposed. The wavelet kernel function is approximately orthogonal and is suitable for the local analysis of signals. In connection with mutual information, Generalized variance as a contrast function to measure the statistical independence, we can get the ideal mathematical properties.The algorithm is applied to a wide range of blind source separation problems and compared with the existing algorithms.Experimental results show that WKGV- KICA algorithm in the same conditions under the separation of higher precision, and better performance.