基于S变换和共空间模式的运动想象脑电特征提取

针对共空间模式算法运用于运动想象脑电信号特征提取分类正确率低、计算实时性差等问题,提出运用S变换结合共空间模式算法对脑电信号进行特征提取方法。经过S变换后的信号具有更加明显的时、频、相特征,再运用共空间模式算法提取特定任务信号成分的特征,最后用支持向量机进行分类。实验结果表明:在S变换采样数较多的情况下,平均正确率达到92.8%,大大超过单纯使用共空间模式算法的正确率。如果降低S变换的采样率,系统实时性得到大幅提升,平均运行时间仅为0.85 s,平均分类正确率可达89.8%,比仅运用共空间模式算法的运行时间缩短30.9%。可见,不仅可提高运动想象脑电信号的分类正确率,还可以提高分类的实时性。

MI-EEG Feature Extraction based on S Transform and Common Spatial Pattern

Aiming at the problems of the low accuracy and low efficiency in commonly used Common Spatial Pattern algorithm, it is proposed that the S Transform be added at the preceding stage to extract the feature of the motor imagination (MI) electroencephalography signal. This paper mainly demonstrates the edge of the S Transform, including the more obvious characteristics of time, frequency and phase. Evidently, with the aid of S Transform to extract the feature of electroencephalography signal both in time and frequency, Common Spatial Pattern could greatly increase the accuracy of classification results. The results of the support vector machine classification illustrates that the accuracy reaches 92.8% when the samples of S-transform are adequate, which is significantly higher than the accuracy of applying Common Spatial Pattern algorithm only. If we qualify the number of samples, the cost of modeling time would only continue 0.85s, 30.9% less than the original method, and could still achieve the accuracy of 89.8%. So, this method is better for feature extraction and both the accuracy and the efficiency of classification are optimized.