fig6

Figure 6. EOG-based human-machine interaction. (A) Schematic illustration of EOG acquisition, signal preprocessing, and deep learning-based command prediction for real-time Pac-Man control; (B-E) Processed EOG signals corresponding to four directional commands (up, left, down, and right); (F) CNN-BiLSTM architecture with an attention mechanism for feature extraction and classification; (G) t-SNE visualization of signal features before and after deep learning-based feature extraction; (H) Normalized confusion matrix for four-class eye-movement recognition; (I) Demonstration of real-time control of a Pac-Man game using EOG signals. EOG: Electrooculography; CNN-BiLSTM: convolutional neural network-bidirectional long short-term memory; t-SNE: t-distributed stochastic neighbor embedding; ACC: accuracy.