Amplification of Crew EEG Signals and Cognitive State Decoding Based on Generative Adversarial Networks

To address the challenge of assessing cognitive load in maritime personnel, this study developed a multi-task navigation simulation paradigm. Fifteen maritime professionals, organized into groups of three, performed relevant cognitive load tasks in a navigation simulator. To overcome the difficulty of limited sample size in EEG data acquisition, this study proposes an Attention-based Long Short-Term Memory Generative Adversarial Network (Attention-LSTM-GAN). This model synthesizes highly biologically plausible multi-band EEG signals and incorporates ship behavioral data as a multimodal validation anchor, thereby establishing a neuro-behavioral synergistic framework for cognitive state assessment. The attention mechanism integrated into the model adaptively enhances the representation of key frequency band features, including theta, alpha, and beta. Experimental results demonstrate optimal performance at 4× data augmentation (θ: 92.59%, α: 86.69%, β: 90%). Multi-band fusion further improves the integrated classification accuracy to 97.78%, significantly outperforming traditional machine learning methods (SVM: 60.49%; CNN: 83.02%; RNN: 82.55%).