摘要
对独立分量分析算法的基本理论和FastICA算法进行了简要介绍.传统的FastICA算法只具有二阶的收敛速度,为了提高独立分量分析算法的收敛速度,减少迭代次数和运行时间,提出了一种改进的独立分量分析算法——五阶收敛的牛顿迭代法.对牛顿迭代算法加以修正,使改进的独立分量分析算法具有五阶的收敛速度.图像信号分离仿真实验表明,改进算法与传统的FastICA算法在分离效果相当的情况下,明显减少了传统的FastICA算法的迭代次数和运行时间,提高了收敛速度和运行效率.
The basic theory of independent component analysis（ICA） and the FastICA algorithm are briefly described.Conventional FastICA algorithm has only a second-order convergence rate,which has to be improved to reduce the iteration steps and running time.An improved ICA algorithm is therefore proposed,i.e.,the Newton＇s iteration method with a fifth-order convergence.It is actually a modified Newton＇s iteration method to enable the improved FastICA algorithm to have fifth-order convergence rate.The simulation results of separating an image signal from others showed that although the algorithm thus improved has the same separating effect as conventional FastICA algorithm,it can greatly reduce the iteration steps and running time further than FastICA,thus increasing the convergence rate and improving the operation efficiency.
出处
《东北大学学报（自然科学版）》
EI
CAS
CSCD
北大核心
2010年第8期1086-1088,1097,共4页
Journal of Northeastern University(Natural Science)
基金
辽宁省自然科学基金资助项目(20072025)
关键词
独立分量分析
定点算法
牛顿迭代法
负熵
最速下降法
independent component analysis
fixed-point algorithm
Newton＇s iteration method
negentropy
steepest descent method
