基于熵的脑电特征选择情绪识别研究

Research on emotion recognition with EEG signal feature selection based on entropy

近年来,利用脑电（EEG）信号识别情绪已经成为一个热门的研究领域。随着科技的发展,脑电采集设备的采样率越来越高,带来了脑电数据量的大幅增加。如何在大量EEG信号中提取并选择最优特征是当前情绪识别面临的一个重要问题。选择脑电信道后进一步提取特征的做法,极易造成有效信息的丢失。为了解决以上问题,提出了一种不需选择通道的情感识别方法。首先,通过经验模式分解将原始脑电信号分解为若干本征模函数（IMF）,并计算本征模函数的样本熵;接着,提出了一种基于熵的特征选择方法,对样本熵进行选择并形成特征向量;最后,将特征向量输入极限学习机（ELM）进行训练和测试。该方法在DEAP数据集上进行了测试,对情绪的平均识别率达88.39%。实验结果表明,该方法能够有效选择特征,对情绪识别具有良好的分类效果。

Emotion recognition by using electroencephalography（EEG） signals has become a prosperous research field in recent years. With the development of science and technology, the sampling rate of EEG acquisition equipment becomes higher and higher, bringing about a large increase in EEG data. Therefore, how to extract and select the optimal features in EEG signals is an important problem in current emotion recognition task. The further features extraction after the selection of the brain channels is very likely to result in the loss of effective information. In order to overcome the difficult points mentioned above, this paper proposes a non-channel selection method for emotion recognition. First, the original EEG signals are decomposed into several intrinsic mode functions（IMFs） by the empirical mode decomposition, and then the sample entropy of intrinsic mode functions are calculated. Second, a feature selection method based on entropy is proposed to select the sample entropy and form the feature vectors. Finally, these vectors are sent to the extreme learning machine（ELM） for Training and test. This method is evaluated on the DEAP database and the average recognition rate of emotion is 88.39%. The experimental results show that this method can effectively select features as well as preferable classification effect on emotion recognition.