Emotion Recognition of EEG Signals Using Wavelet Filter and Convolutional Neural Networks

EasyChair Preprint 6623

Abstract

Emotion is a psychophysiological process that is triggered by conscious or unconscious states. Emotional information in the human brain can be captured through a multi-channel Electroencephalogram (EEG). EEG signals are recorded from multiple channels, representing information points of electrical activity from different parts of the brain. While the EEG signal of each channel is a sequence, some studies use one dimension in recognizing patterns, and the signal from the next channel is a continuation of the sequence from the previous channel. It makes the channel sequence less maintained so that the EEG signal processing from multi-channel is seen as a matrix, i.e., the vertical direction is the signal from various channels. While the horizontal direction of the sequence of each channel. So that the signal processing of the multi-channel is rich in information in the appropriate order, this study used 2D Convolutional Neural Networks (CNN) for emotion recognition, with various architectures and configurations to get the best performance. In addition, the EEG signal needs to be extracted, which reflects the emotion variable first using a Wavelet. That is the 4-45 Hz frequency band of Theta, Alpha, Beta, and Gamma. The results show that two-dimensional CNN, which pays attention to signal order, produced the best accuracy of 83.44% compared to 75.97% with one-dimensional CNN. Experiments gave the best configuration used eight layers and Stochastic Gradient Descent (SGD) weight correction.

Keyphrases: CNN, EEG signal, Wavelet, emotion recognition, multi-channel

@booklet{EasyChair:6623,
  author    = {Rofiqoh Hadiyan Dien Haqque and Esmeralda Contessa Djamal and Arlisa Wulandari},
  title     = {Emotion Recognition of EEG Signals Using Wavelet Filter and Convolutional Neural Networks},
  howpublished = {EasyChair Preprint 6623},
  year      = {EasyChair, 2021}}