Multi-Task Improve Domain Generalization in EEG-based Emotion Classification using Feature Fusion Learning Model

The use of Electroencephalography (EEG) signals for emotion identification tasks is common. However, the domain shift issue may cause EEG-based emotion detection models' performance to decline when used in new domains. In this article, we present a feature fusion learning model-based multi-task learning strategy to enhance domain generalization in EEG-based emotion classification. By simultaneously learning a shared feature extractor and numerous domain-specific classifiers, we train a single model to categorize emotions across various domains. Additionally, we present a feature fusion module that combines time-domain, frequency-domain, and wavelet-domain EEG characteristics. We run trials on the DEAP and SEED EEG emotion recognition datasets to assess our suggested methodology. The experimental findings show that our suggested strategy outperforms a number of cutting-edge techniques in terms of classification accuracy, both within- and across-domains.