CSP- LSTM Based Emotion Recognition from EEG Signals

Abstract

Emotion recognition from the electroencephalogram (EEG) signals is a growing area of research in the field of neuroscience and medical technology. The ability to accurately identify and classify emotions can have significant implications for various applications, including the diagnosis and treatment of emotional disorders, human-computer interaction, etc. A deep learning-based emotion identification method is proposed here. The EEG data is divided into four frequency bands (alpha, beta, lower gamma and upper gamma) using a temporal domain filter, and then the common spatial pattern (CSP) feature extraction is applied to obtain a set of discriminative features that are used to train the long short-term memory (LSTM) for emotion classification. The LSTM was trained for each band and the predictions of each of these models (same architecture with different weights) were combined using a majority voting classifier (MVC) method. All subjects achieved accuracy values exceeding 90%, resulting in a mean accuracy of 97.34% when considering the average across all subjects. The utilisation of this method for bio-signal modeling and interpretation has the capability to enhance the identification and management of emotional disorders. etc.