A Hybrid Approach for Emotions Classification using Modified Intrinsic Time Scale Decomposition

Emotion recognition is a crucial area of research in brain-computer interactions (BCI). The swift progress of BCI has established emotion identification through electroencephalogram (EEG) data as a prominent area of research in recent years. The intricacy and variety of emotional signals make EEG-based emotion recognition a difficult task in identifying patterns. To address this issue, this study develops an automated classification model utilizing the PYNQ-Z2 platform for the recognition of human emotions. This model employs a modified intrinsic time scale decomposition to analyze the complicated EEG signal by breaking it down into numerous proper rotational components (PRC). The ideal PRC is subsequently chosen and input into the Convolutional Neural Network for automated feature extraction. The feature set is subsequently assessed using three distinct classifiers: Quadratic Discriminant Analysis, Multilayer Perceptron Network, and Ensemble Voting Classifier. The study utilized the DEAP dataset and achieved accuracies of 94.11%, 94.70%, and 94.76% for the valence, arousal, and dominance domains, respectively. The proposed strategy exhibits enhanced efficacy relative to alternative existing alternatives. Additionally, the model is executed on the PYNQ-Z2 board, demonstrating that the proposed technique is compatible with hardware.