【作者】 李洋；

【导师】 胡斌；

【作者基本信息】 兰州大学 ， 信息与通信工程·通信与信息系统， 2016， 硕士

【摘要】 脑电图作为一种非倾入式的记录脑活动信息的方法,由于其高分辨率的特性而被广泛用于研究大脑功能。由于脑电信号非常微弱,极易受到其它噪声伪迹的干扰,尤其是眼电伪迹,而这严重影响了脑电的后续分析。因此,如何有效地去除脑电中的噪声并获得干净的脑电信号成为了一项重要的研究内容。目前,各种去眼电伪迹的方法层出不穷,但是许多方法受到各种条件的限制,例如很多方法去除眼电伪迹的时候需要同步记录的眼电信号。而在便携式环境下,由于脑电采集设备的导联数目受到限制,因此这些方法并不能直接进行使用。为了解决上述问题,本文提出了小波变换结合卡尔曼滤波器模型用于自动地去除脑电中的眼电伪迹。由于眼电伪迹是随机出现在脑电信号中的,所以该模型首先检测并提取眼电伪迹出现的区域,并只对这些区域进行处理从而避免对非眼电区域的脑电信号造成影响。然后用小波将提取出来的眼电区域分解为4层,由于眼电信号的频率较低,所以只用近似层的小波系数重构出粗糙的眼电伪迹。因为重构出的眼电伪迹中含有低频的脑电信号,因此采用卡尔曼滤波器对重构的眼电伪迹做进一步处理,估计出纯粹的眼电伪迹,最后从原始的脑电信号中将其去除,即可得到干净的脑电信号。为验证上述模型的有效性,本文使用模拟数据将该模型和ICA结合小波模型及自适应滤波器结合小波模型进行了定量的对比分析。在时域上,本文所提方法的均方误差(MSE)值为0.0017,而ICA小波模型和自适应滤波器小波模型的MSE分别为0.0468和0.0052;在频域上,本文所提方法的平均绝对误差(MAE)的均值为0.0052,而另外两种模型的MAE的均值分别为0.0218和0.0115。可以看出,本文所提模型的MSE值和MAE的均值都要明显小于另外两种模型,说明该模型在时域上和频域上都取得了更好的去眼电伪迹效果。为进一步说明该模型的实用性,本文还分别使用了由BrainCap和三导脑电采集器采集的实际脑电数据对该方法的去噪性能加以验证,结果表明该方法能很好地去除脑电中的眼电伪迹。综合来看,本文提出的方法能够有效地自动去除脑电中的眼电伪迹,而且能够很好地应用于便携式脑电设备。更多还原

【Abstract】 Electroencephalogram(EEG) is a non-invasive method to record electrical activity of brain and it has been used extensively in research subjects of brain function due to its high time resolution. The EEG is so weak that it is always interfered with other noises, especially the ocular artifacts, which makes the subsequent analysis of EEG hard. Hence, how to eliminating ocular artifacts from EEG becomes the current research hotspot. Now, although there are many methods of ocular artifacts removal, lots of them are limited by various conditions, for example, some methods need the synchronous electrooculorgram when they are used to eliminate ocular artifacts. Since the channels’ number of the portable EEG-recording device is finite, so these methods are not used directly.In order to solve the above-mentioned problem, we have proposed a new model combined by wavelet transform and Kalman filter for automatically eliminating ocular artifacts. Since the artifacts appear randomly in the EEG, so this method detect artifacts areas and just deal with them to avoid impacting on EEG out of these areas. Then, wavelet is employed to decompose artifacts areas into 4 layers and the approximation coefficients is used to construct coarse ocular artifacts due to the electrooculogram mainly concentrating in the low frequency band. Because the ocular artifacts contains low-frequency EEG, Kalman filter is applied to process the constructed ocular artifacts to estimate the pure artifacts, which are subtracted from the raw EEG to acquire clean EEG at last.We compare the proposed model with ICA-wavelet model and adaptive-filter-wavelet model using simulated data to test effectiveness of the new method. In time domain, the Mean Squared Error(MSE) of the new method is 0.0017, while the MSE of the other two models are 0.0468 and 0.0052, respectively. In frequency domain, the Mean Absolutely Error(MAE) using the new method achieved an average of 0.0052, the average of MAE of the other two model are 0.0218 and 0.0115, respectively. Obviously, both MSE and average of MAE of the new model are significantly lower those of the other two models, which indicate our proposed model achieved better result of ocular artifacts removal. To further test practicability of the new model, we use it to eliminate ocular artifacts in EEG which are recorded by BrainCap and three-channel EEG collector, respectively, it turns out that the proposed model can remove artifacts effectively. Taken together, the new model can eliminate ocular artifacts in EEG automatically, and it is recommended for portable applications.更多还原