论文部分内容阅读
准确对事件诱发电位(ERPs)进行分类,对于各种人类认知研究和临床医学评估非常有意义.由于ERPs信号是非常高维的数据,而且其中包含非常多的与分类无关的信息,从ERPs信号中提取特征尤显重要.分析了共空间模式(CSP)的原理和不足,引入自回归(AR)模型与白化变换相结合,提出了针对ERPs分类的时空特征提取方法,并设计了验证该方法的认知实验,在认知实验数据上分别用时空特征提取方法与CSP提取特征,用同样的分类器支持向量机(SVM)训练分类器,比较它们的分类效果.实验表明,在ERPs分类问题上,时空特征提取方法与CSP相比具有明显的优势,在参数确定合理的情况下,时空特征提取方法可使分类准确率达到90%以上.
Accurate classification of Event Evoked Potentials (ERPs) is significant for a variety of human cognitive and clinical evaluations.As ERPs are very high-dimensional data and contain a large number of information that is not relevant to the classification, It is very important to extract the features from the signal.Analysis of the principle and insufficiency of CSP and the combination of autoregressive (AR) model and whitening transform, the paper proposes a method of spatio-temporal feature extraction for ERPs classification, Method in cognitive experiment, extract features from spatio-temporal feature extraction method and CSP separately on cognitive experiment data, and train the classifiers by the same SVM to compare their classification results.Experiments show that in the classification of ERPs Problem, the spatio-temporal feature extraction method has obvious advantages over CSP. Under the condition of reasonable parameters, the spatio-temporal feature extraction method can make the classification accuracy rate more than 90%.