EEG-Based Emotion Classification By Using Convolutional Neural Network (CNN)

Tong, Siau Khee (2017) EEG-Based Emotion Classification By Using Convolutional Neural Network (CNN). Project Report. Universiti Teknikal Malaysia Melaka, Melaka, Malaysia. (Submitted)

Text (24 Pages) EEG-Based Emotion Classification By Using Convolutional Neural Network (CNN).pdf - Submitted Version Download (655kB)

Abstract

Emotion is an essential component in social interaction and communication which act as a guideline in decision making or planning.It is believed that emotion can help to bridge the gaps between human and computer.However, emotion detection through facial expression or speech detection that existing nowadays is not accurate enough because it can be faked.Therefore,a EEG-based emotion classification system is proposed to analyse the “inner” emotion through the brain activity.Convolutional neural network (CNN) technique is applied to perform feature extraction and classification for the labelled EEG signal.This method aims to replace the expensive traditional hand-crafted feature extraction method.Based on the spatial temporal characteristics of EEG signal, two CNN architectures are built by referring to the previous researches.It was tested with EEG dataset that acquired from 32 subjects by using pictures from IAPS database as stimuli and verified by using SEED dataset. The system able to achieve an average accuracy of 75% in classifying 3 emotions (positive,neutral,negative) from SEED dataset.However,it only able to achieve average accuracy of 30% for 4 class of emotions and 59% for 2 class of emotions from self-conducted dataset due to the insufficient of training data.

Actions (login required)

View Item