字典训练结合形态分量分析的诱发电位少次提取方法  

Few-trial Extraction of Evoked Potentials with Dictionary Training and Morphological Component Analysis

作　　者：毕峰[1,2] 邱天爽[1] 余南南[1] 

机构地区：[1]大连理工大学电子信息与电气工程学部,大连116024 [2]辽东学院信息技术学院,丹东118000

出　　处：《信号处理》2013年第3期405-409,共5页Journal of Signal Processing

基　　金：国家自然科学基金资助项目(61172108;61139001;81241059);国家科技支撑计划项目(2012BAJ18B06)

摘　　要：诱发电位的少次提取对于研究大脑活动规律以及临床诊断等均有重要意义。根据诱发电位与自发脑电信号的不同特点,本文提出一种基于形态分量分析的诱发电位少次提取方法,在不同的过完备字典上对诱发电位与自发脑电信号进行稀疏表示。为了改善在稀疏表示过程中的错误分解问题,提出使用几次带噪观测信号的叠加平均结果作为模板信号,并使用K-SVD算法训练得到合适的过完备字典,再对当前观测信号进行混合稀疏表示。实验结果表明,该方法能够有效地降低由通用过完备字典进行稀疏表示时的错分程度,较好地实现对诱发电位信号的提取。The few-trial extraction of evoked potentials is very meaningful to the study of brain and many clinical applica- tions. In this paper, we proposed a few-trial extraction method based on the morphological component analysis. That is, the evoked potential and the electroencephalogram were sparsely represented in the different overcomplete dictionaries. To avoid the error representation due to the selection of inappropriate dictionaries, we used the average resuh of several noisy signals as the template signal and employed the K-SVD algorithm to obtain the appropriate overcomplete dictionaries in accordance with different signals, and then sparsely represented the corresponding signals in these trained dictionaries. Experimental re- suits show that the algorithm can reduce the inappropriate representation efficiently versus the method with the universal overcomplete dctionaries, and it can extract the evoked potentials better than the latter.

关 键 词：诱发电位提取 形态分量分析 字典训练 K—SVD 

分 类 号：TN911.7[电子电信—通信与信息系统]