EEG-based emotion classification using convolutional neural network

Abstract

Electroencephalograph (EEG) signals can real-time reflect the brain activity. Using EEG signal to analysis human emotional states is a common research. Brain network analysis is a way to study brain emotional activity, it bases on the graph theory and finds the brain connectivity patterns. This way should calculate the matrices of functional connectivity of EEG and extract the characteristics from the matrices. This paper describes a straightforward way to use the matrices of functional connectivity and extract feature by using Convolution Neural Network (CNN). CNN was employed to accomplish several task: 1) 2-classification task, 2) 3-classification task and 3) 4-classification task, and the average accuracy of 2-classification task is about 85%, 3-classification task is about 78% and 4-classification is about 75%. The study demonstrations that the matrices of functional connectivity carries important informations about the emotional states, and the CNN model can extract the distinguishing featurse to differentiate the emotional states.