Research On Emotion Recognition Based On EEG

Emotion is closely related to daily life,individuals that in positive emotions can keep good working statues,and negative emotions can weaken the attention and judgement of people,what’s more,it would have influence on the mental and physical health,and even damage their life and property if the situation is ignored.Meanwhile,the emotion based Human-Computer Interaction is the key to the development of Artificial Intelligence,take all this conditions into consideration,it prove that the accurately and rapidly identify the emotion state of subject has a certain practical significance.Electroencephalogram(EEG)is the direct production of brain activity,which has a close relationship with emotion,the emotion recognition of EEG signals has become one of the hot issues that attracted scholars’ attention.At present,the problem of EEG emotional recognition is the results of some subjects are inaccurate,it is because of the mismatch between individual specificity and global threshold.To solve this problem,ReliefF matching feature selection algorithm(RMFS)is proposed in the thesis,by selecting the feature types and channels,by using this algorithm,it can effectively obtain the matching features and eliminate the irrelevant information and redundant information of single subject,the problem of the individual specificity is solved.The experimental process based on RMFS algorithm is shown as follow: Firstly,a5-layer DB5 wavelet packet decomposition is performed on the selected EEG signals from DEAP database,and the signals are reconstructed into six emotion-related bands;secondly,the Empirical Mode Decomposition(EMD)is used to obtain the Intrinsic Mode Function(IMF)components of each band,and the ten types of feature respectively based on wavelet coefficients and reconstructed signal IMF components are extracted;thirdly,the types of feature and channels are selected by RMFS algorithm so that obtain the matching feature set and it’s weights;finally,by utilizing Support Vector Machine(SVM),emotion are recognized as happiness,anger,sadness,and relax,and recognition accuracy are collected.Though the feature selection experiment of the individual subjects,the improvement of computing efficiency is tested;by the analysis of the construction of the matching feature set,the four main features of overall recognition are collected,and the necessity of utilize the RMFS algorithm is proved.And there are 3 kinds of emotion recognition method are utilized in this thesis,they are: the emotion two-category of valence andarousal,the emotion four-class classification,and the recognition of each two emotion respectively,according to the experimental result,the effectiveness of RMFS is verified.Meanwhile,the fact that RMFS algorithm has more advantages compared with the tradition method in the muti-class classification condition is found out.There are 23 figures,6 tables,and 68 references in the thesis.