基于奇异值第一主成分的睡眠脑电分期方法研究

目的:脑电信号含多种噪声和伪迹,信噪比较低,特征提取前必须进行复杂的预处理,严重影响睡眠分期的速度。鉴于此,本文提出一种基于奇异值第一主成分的睡眠脑电分期方法,该方法抗噪性能较强,可省去预处理过程,减少计算量,提高睡眠分期的效率。方法:对未经过预处理的睡眠脑电进行奇异系统分析,研究奇异谱曲线,提取奇异值第一主成分,探索其随睡眠状态变化的规律。并通过支持向量机利用奇异值第一主成分对睡眠分期。结果:奇异值第一主成分不仅能表征脑电信号主体,而且可以抑制噪声、降低维数。随着睡眠的深入,奇异值第一主成分的值逐渐增大,但在REM期处于S1期和S2期之间。经MIT-BIH睡眠数据库中5例同导联位置的脑电数据测试(仅1导脑电数据),睡眠脑电分期的准确率达到86.4%。结论:在未对脑电信号进行预处理的情况下,提取的睡眠脑电的奇异值第一主成分能有效表征睡眠状态,是一种有效的睡眠分期依据。本文运用提出的方法仅采用1导脑电数据,就能得到较为满意的睡眠分期结果。该方法有较强的分类性能,且抗噪能力强,不需要对脑电作复杂的预处理,计算量小,方法简单,很大程度上提高了睡眠分期的效率。

Study of Sleep EEG Staging Method Based on the First Principal Component of Singular Value

Objective： EEG signal contains a variety of noise and artifacts, it has low SNRI The complex pre-processing must be done before the feature extraction. It will seriously affect the speed of the sleep staging. In view of this, a sleep staging method based on the first principal component of singular value is proposed in this paper. This method is robust to noise. The need for pre-processing is eliminated to reduce the amount of computation and improve the efficiency of sleep staging Methods： By singular value decomposition （SVD） on the EEG without pre-processing, study the singular spectrum curve. And extracting the first principal component of singular value on EEG to explore the rule with the change of sleep states. The SVM is used for sleep EEG stage determination. Results： The first principal component of singular value can not only characterize the sleep states, but also can restrain noise and reduce dimension. Along with the deepening of sleep, the first principal component of singular value gradually increases its value, but between S1 and $2 in REM. Be tested on the 5 cases EEG data with the same channel position in the MIT-BIH database （one channel data only）, the accuracy that using the proposed scheme achieves 86.4%. Conclusion： In the case of no EEG pre-processing, the first principal component of the singular value of the sleep EEG can effectively characterize the sleep states. It is an effective basis for sleep staging. In this paper, only one channel EEG data is used, and the proposed scheme can get a satisfactory sleep staging result. This scheme can provide a strong classification performance, and it has the advantages of restrain noise, which requires no complex preprocessing on EEG. Therefore, small amount of calculation is needed. It is a simple method that can greatly improve the efficiency of sleep staging.