基于遗传算法的运动想象脑电信号分类准确率的提升方法

An improved classification method for motor imagery EEG signals based on genetic algorithm

为了提高运动想象脑电信号的分类准确率,本研究提出了一种基于遗传算法(genetic algorithm,GA)的脑电信号分类提升方法。该方法利用遗传算法与共空间模式算法(common spatial pattern,CSP)相结合,进行不同时间段的特征提取,再利用遗传算法得到不同时间段对分类正确率的权值及数据可信度。利用本实验室采集的脑电信号进行测试,分类准确率由加权前的80%左右提升至加权后的95%以上。实验结果证实,该方法可以有效提高脑电信号分类准确率,并且可以根据可信度剔除低质量的数据。同时,该方法还可以与其他特征提取方法相结合,对不同时、频特性进行有效性及可信度计算,提升分类准确率。这也是本方法更深一层的意义。

In order to improve the recognition rate of motor imagery EEG signals,an improvement classification method based on genetic algorithm（ GA） was proposed. The proposed method combined GA and common spatial pattern（ CSP） to extract the features of different time. After considering the classification accurate,GA was used to calculate different time slices＇ weights. And based on the weights,the data credibility was calculated. Using the EEG signals collected in this laboratory,the accuracy of classification improved from about 80% before weighting to more than 95% after weighting. The experimental results confirm that this method can effectively improve the classification accuracy of EEG signals,and can eliminate low-quality data according to credibility. At the same time,this method can also be combined with other feature extraction methods to calculate the validity and credibility of different time and frequency characteristics to improve the classification accuracy.