基于mRMR特征优选与SVM的脑电信号的情绪识别

Emotion Recognition of EEG Signals Based on mRMR Feature Optimization and SVM

脑电信号情绪识别的难点之一在于如何优选特征,为此,本文提出一种基于mRMR特征优选方法的脑电信号情绪识别。首先,对数据集中经预处理后的脑电信号提取时域、频域、非线性、Hjorth特征等多种特征,构建特征向量;其次,采用mRMR优选特征向量数量,获得最优特征向量组;最后,采用支持向量机算法(SVM)对优选的特征进行分类,获得效价和唤醒度的二分类结果。实验结果表明,基于mRMR特征选择方法的脑电信号情绪识别方法在效价和唤醒度两个方面的识别准确率为76.82%和77.6%,具有良好的识别效果。

One of the difficulties in emotion recognition of EEG signals is how to optimize features.Therefore,this paper proposes a method for emotion recognition of EEG signals based on mRMR feature optimization.Firstly,extract various features such as time-domain,frequency-domain,nonlinear,and Hjordh features from the preprocessed EEG signals in the dataset,and construct feature vectors.Secondly,mRMR is used to optimize the number of feature vectors and obtain the optimal set of feature vectors.Finally,support vector machine(SVM)algorithm is used to classify the selected features and obtain binary classification results of potency and arousal.The experimental results show that the emotion recognition method based on mRMR feature selection for EEG signals has recognition accuracy of 76.82% and 77.6% in terms of potency and arousal,and has good recognition performance.