一种基于十五阶的FastICA改进算法

快速独立分量分析（FastICA）因其收敛速度快而被受关注，但存在初始值选取不当可能导致算法的收敛速度减慢甚至不收敛的问题。针对基本牛顿迭代FastICA算法对初始值选择比较敏感的缺点，以最大化负熵为目标函数，引入十五阶牛顿迭代的修正形式对FastICA算法的核心迭代过程进行改进，改进算法的收敛性不再依赖于初始值的选择，而且具有更快的收敛速度。将改进算法应用到仿真实验，实验结果显示，改进算法在分离效果相当的前提下，迭代次数更少，收敛速度更快，而且收敛速度更加稳定。

Improved Fast ICA algorithm based on fifteen-order Newton iteration

Fast Independent Component Analysis（FastICA） has attached broad attention as its fast convergence. However, if the initial vectors are chosen incorrectly, the algorithm may converge slowly or even not converge. The FastICA algorithm based on the-fifteen order Newton iterative correction form is improved by modifying kernel iterative process aiming to solve the problem that basic FastICA algorithm is sensitive to initial vectors. The improved algorithm convergence is independent on the initial values, and has a faster convergence speed compared with the basis FastICA. Simulation experiments show that the iteration number decreases with more stable convergence speed compared with the basis FastICA.