Adaptive Spectral Correlation Convolutional Neural Network for EEG-Based Emotion Recognition: A Study on the SEED Dataset

This research presents the Adaptive Spectral Correlation Convolutional Neural Network (ASCCNN), an advanced framework tailored for precise emotion recognition using EEG (Electroencephalography) data. ASCCNN integrates adaptive spectral correlation techniques into its architecture, enhancing sensitivity to nuanced frequency patterns indicative of emotional states. Utilizing the SEED IV dataset, the study employs sophisticated preprocessing methods such as band differential entropy and baseline removal to refine EEG features. These processed features are then fed into ASCCNN's specialized convolutional layers designed to capture intricate spectral correlations critical for emotion classification. ASCCNN's training regimen involves fine-tuning model parameters using optimization techniques like the Adam optimizer with carefully chosen learning rates. The training process is meticulously monitored to ensure convergence and prevent overfitting, guided by validation metrics for model refinement. Post-training, ASCCNN undergoes rigorous evaluation on a dedicated test set to gauge real-world performance in emotion recognition tasks. Evaluation metrics encompass accuracy, loss, and spectral correlation analyses, offering comprehensive insights into the model's effectiveness. The ASCCNN achieved a notable accuracy of 96.15%, demonstrating its robustness and precision. The study concludes with visualizations of training and validation metrics, showcasing ASCCNN's adaptive learning prowess and its ability to discern subtle spectral correlations essential for accurate emotion recognition in EEG data. ASCCNN represents a significant advancement in neuroinformatics, providing a robust framework for emotion analysis with potential applications in affective computing and neuroscience research.