论文部分内容阅读
目的:眼电是脑电的主要干扰,采用眼电信号作为参考的自适应滤波能有效消除眼电干扰。然而眼电采集不方便且繁琐。在脑机接口中为从脑电中去除眼电,提出基于约束独立分量分析和自适应滤波的快速去除方法。该方法具有无需记录眼电信号且快速的优点。创新点:所提方法避免了实验过程中直接对被试者进行眼电信号采集,减少被试者在实验过程中的不适。该方法处理后的识别正确率比单纯用传统ICA算法和不进行任何处理的源信号分别提高了3.3%和12.6%。另外,该方法的时间耗费较上述两种算法分别降低了83.5%和83.8%,更好地满足脑机接口在线要求。方法:该方法分为两个阶段:第一阶段的目的是提取纯净的EOG信号。首先用ICA算法将输入信号分离成相互独立的分量(IC)。计算每个IC的峰态系数值并依据该值自动识别EOG独立分量(图2)。然后运用经验模态分解(EMD)将所识别的EOG信号自适应分解成数个IMF。根据IMF频域特征,选择数个IMF组合成纯净的EOG信号(图3)。第二阶段的目的是结合SCICA和RLS滤波算法去除混合在EEG信号中的EOG伪迹。首先SCICA利用第一阶段分离出的纯净EOG信号作为参考模板,迅速将混合在源信号中的EOG信号识别分离出。然后将该EOG信号分量作为RLS滤波器参考信号进行自适应滤波,最终去除EOG伪迹(图7)。结论:针对脑机接口脑电信号包含的眼电伪迹,提出一种基于约束独立分量分析和自适应滤波的快速自动去除方法。该方法去除效果良好,可用于脑机接口中眼电的在线自动消除。
Objective: EEG is the main disturbance of EEG, the use of EEG signals as a reference adaptive filtering can effectively eliminate ocular interference. However, ocular acquisition is not convenient and cumbersome. In the brain-computer interface, EEG was removed from EEG, and a fast method based on constrained independent component analysis and adaptive filtering was proposed. This method has the advantage of no need to record ocular signals and is fast. Innovative point: The proposed method avoids direct acquisition of eye subjects during the experiment, reducing the subject’s discomfort during the experiment. The accuracy of this method is 3.3% and 12.6% higher than the traditional ICA algorithm and the source signal without any processing. In addition, the time cost of this method is reduced by 83.5% and 83.8% respectively compared with the above two algorithms, which can better meet the online requirements of brain-computer interface. Method: This method is divided into two stages: The first stage is to extract the pure EOG signal. First, the ICA algorithm is used to separate the input signal into independent components (ICs). Calculate the kurtosis coefficient value for each IC and automatically identify the EOG independent component based on that value (Figure 2). Then the EMD is used to decompose the identified EOG signal into several IMFs adaptively. According to the IMF frequency domain features, several IMFs are selected to form a pure EOG signal (Figure 3). The purpose of the second phase is to remove the EOG artifacts mixed in the EEG signal by combining the SCICA and RLS filtering algorithms. First of all, SCICA uses the pure EOG signal separated in the first phase as a reference template and rapidly separates the EOG signal mixed in the source signal. The EOG signal component is then adaptively filtered as an RLS filter reference signal, eventually removing EOG artifacts (Figure 7). Conclusion: According to the electro-oculogram artifacts contained in brain-computer interface (EEG) signals, a fast and automatic method based on constrained independent component analysis and adaptive filtering is proposed. The method has good removing effect and can be used to automatically eliminate the eye electricity in the brain-computer interface.