一种改进的FastICA算法及其应用.pdfVIP

第 28卷第 4期 计算机应用 V ol. 28 No. 4 2008年 4月 ComputerApp lications Apr. 2008 : 1001- 9081( 2008) 04- 0960- 03 FastICA 郭 武, 朱长仁, 王润生 ( ATR , 410073) (guouw u2000@ ya oo. com. cn) : 独立分量分析是基于信号高阶统计量的信号分析方法, 它可以找到隐 在数据中的独立分量, 已经广泛 应用到语音信号处理图像处理及信息通信等方面目前应用较多的快速独立分量分析( FastICA)利用了牛顿迭代法 原理, 具有较快的收敛速度, 但对初始值的选择比较敏感为克服其缺点, 改进其优化学习算法, 在牛顿迭代方向增加 一维搜索,使改进后的算法的收敛性不依赖于初始值的选择将改进的 FastICA算法应用到运动目标检测中, 取得稳 定性较强的结果 : 独立分量分析; 快速独立分量分析; 运动目标检测 : TP391. 4 1 : A An im proved FastICA algorithm and its app lication GUO Wu, ZHU C ang-ren, WANG Run-s eng (A TR Labora tory, N ationa l Un iversity of D ef ence Techn ology, Chang sha H unan 410073, China ) A bstract: Independent Com ponent Analysis ( ICA) is a signal analysism et od based on ig order cum ulants of signals and it can find out t e latent independent com ponents in data. Recently ICA as been w idely used in m any fields suc as speec recogn ition, mi age processing, telecommunication system etc. T e FastICA is t e m ost popu lar algorit m for ICA at present, and it uses New ton ru le to optmi ize t e objective function. T is algorit m can converge speedily bu t is not robust to in itialization. In order to overcom t e drawbacks, one dmi ens ion searc w as mi posed on t e d irection of New ton iterative. T e mi proved algorit m can ensure t e convergence of t e resu lts and is robust to in itialization. W en t e mi proved algorit m is used to detect t e m oving target, t e expermi ental resu lts s ow t at it is a robust m et od. K ey w ords: Independent Component Analysis ( ICA ); FastICA; detection of moving target