Enhanced Emotion Recognition from EEG Signals with ResConv U-Net Architecture

Despite the progress in the ability to detect emotions from EEG-based signals, there is still a growing need for human-computer interaction. As emotions are subjective and influenced by several variables, recognizing them is more challenging. In recent years, subsequences of the neural network have been widely utilized to analyze the signals gained from the brain. This study presents a novel CNN architecture that combines the strengths of the residual learning and U-Net upsampling strategies, designed for one-dimensional signal analysis. An encoder-decoder structure in the proposed model effectively captures and reconstructs important temporal features of the EEG signals. The model was applied to the DEAP dataset, and evaluation metrics such as F1-score, recall, and precision were calculated and achieved accuracy in arousal and valence showing the priority of the proposed model.