Feature Extraction And Analysis Of Emotional Pictures Visual Evoked EEG

In 1872, Darwin wrote a book called "the Expression of the Emotion in man and Animal", in which he said that emotion is an adaptation tool in advanced evolutionary stage, since then people started emotional experiments and theoretical research. More than 100 years later, to the late 20th century, Emotional research flourished and combined with cognition, neuroscience and brain science et al; the research methods are various, such as Electroencephalography (EEG), functional Magnetic Resonance Imaging (fMRI ), functional Near Infrared Imaging (fNIRI) and so on. Because of the high temporal resolution, low cost and convenient, EEG is widely used in emotional research.This thesis designed an emotional pictures evoked experiment based on International Affective Picture System (IAPS). In the experiment, subjects viewed emotional pictures of different emotional levels and acquired EEG signals at the same time. Found the EEG feature associated with emotional changes by feature extraction and analysis of EEG signals, and tried to establish the corresponding relationship between the emotional levels and EEG feature, and classified emotional levels according to EEG feature. First, the thesis calculated EEG power spectrum and painted its brain mapping. The brain mapping showed emotional images evoked EEG is most active in the prefrontal area. EEG spectral analysis showed that the signal energy is mainly concentrated in the 15Hz or less. In order to find the most discriminative frequency of emotional evoked EEG, the thesis analyzed the discriminative frequency of some channels. Then, this thesis analyzed the power spectral entropy and correlation dimension of EEG, and used least-squares fitting to establish the corresponding relationship between the emotional levels and EEG feature in channel AF3, AF4, F3, F4. In the pattern recognition part, used 5-fold cross-validation Support Vector Machines and Hidden Markov Model to classify and recognize EEG features. In order to improve classification accuracy rate, the thesis fused the feature at feature level, and then used the same methods to classify and recognize the fused feature.The results showed that after the feature fusion, the highest average classification accuracy rate reached 86.5% of classifying the emotional pictures of emotional level one, five and eight. It's able to classify most negative, neutral and most positive emotional state more objectively by emotional pictures evoked EEG; the next step will research new feature extraction and classification methods to distinguish each emotional level.