A Transfer Learning Approach for the Classification of Human Emotions Using Electroencephalogram Signals

In this research, an attempt to develop an emotion recognition system is made using electroencephalography (EEG) signals. Firstly, discrete wavelet transform is applied to the EEG signals and is decomposed into its alpha, beta, gamma, delta, and theta bands. Secondly, principal component analysis is implemented on the extracted features to make them mutually uncorrelated preserving their dimensionality. Finally, the extracted features are then classified using a pre-trained transfer learning model such as MobileNetV3. The proposed method attained an accuracy of 87.51% on a standard DEAP dataset which is comparable with advanced modern methods for emotion recognition.