Keywords:Emotion Recognition, Electroencephalogram, Convolutional Neural Network
In this research, we focus on the role of constraints introduced from the psychophysiological studies to emotional recognition using EEG (Electroencephalogram) and Deep Learning. We especially focus on feature extractions using CNN (Convolutional Neural Network) by applying the frequency analysis methods for EEG. In the experiments, our method showed the possibility of classifying positive, neutral and negative emotional states from the features of EEG frequency analysis, and it was also shown that simple constraints like reducing input features are not always effective. Although our proposed method did not exceed the performance of a standard machine learning method, visualization methods of CNN reveal important components relate to the recognition of target emotion from frequency, spatial and temporal axes of EEG. Summarizing the results, it was suggested that our method is not only useful for developing classifiers but also effective to analyze the relationships of EEG and emotion states.