Published Paper Details:

Emotion recognition from electroencephalography (EEG) signals has become a pivotal aspect of affective computing. This research proposes the concatenation of two novel deep neural network architectures to advance the state-of-the-art in EEG-based emotion classification. The first model termed Hybrid STANN with Graph-Smooth Signals, employs a unique combination of spatiotemporal encoding and recurrent attention network blocks. Graph signal processing tools are applied as a preprocessing step for spatial graph smoothing, enhancing the interpretability of physiological representations. The model outperforms existing methods on the DEAP dataset for emotion classification. Additionally, its robustness is demonstrated through successful transfer learning from DEAP to DREAMER and the Emotional English Word (EEWD) datasets, showcasing its effectiveness across diverse EEG-based emotion classification tasks. The second model, named 3DCANN: Spatio-Temporal Convolution Attention Neural Network, addresses the dynamic nature of EEG signals in emotional states. The 3DCANN model features a spatiotemporal feature extraction module and an EEG channel attention weight learning module. By effectively capturing the dynamic relationships and internal spatial relations among multi-channel EEG signals, the model surpasses state-of-the-art performance on the (SEED) Dataset. The integration of dual attention learning and SoftMax classification enhances the model's ability to discern intricate patterns in EEG signals, resulting in superior emotion recognition accuracy. Both proposed models contribute to EEG-based emotion recognition by introducing innovative architectural elements and demonstrating their efficacy through comprehensive evaluations of diverse datasets. This research opens avenues for further exploration in physiological data-driven affective computing applications.