Application of the CNN-BiGRU-Attention Network Optimized by Secretary Bird Optimization Algorithm in EEG-Based Emotion Recognition

Electroencephalogram (EEG) signals, as a real-time and non-invasive physiological modality, have been widely employed in intelligent perception tasks such as emotion recognition. To address the limitations of existing emotion recognition models in spatiotemporal feature extraction and hyperparameter optimization, this study proposes a hybrid deep neural architecture that integrates Convolutional Neural Networks (CNN), Bidirectional Gated Recurrent Units (BiGRU), and an attention mechanism. Moreover, a newly proposed metaheuristic algorithm, the Secretary Bird Optimization Algorithm (SBOA), is introduced to perform global optimization of key hyperparameters. Inspired by the secretary bird's dual-stage hunting and evasion behavior in nature, SBOA exhibits strong exploration and exploitation capabilities. Experiments were conducted on the publicly available DEAP EEG dataset for a four-class emotion classification task. Results demonstrate that the SBOA-optimized model significantly outperforms conventional non-optimized counterparts in terms of classification accuracy, robustness, and convergence speed, achieving a four-class accuracy of 88.9%, which surpasses comparative models based on Genetic Algorithm (GA) and Particle Swarm Optimization (PSO).