| dc.description.abstract | The process of transmitting signals from the brain to execute actions can be broadly divided into three stages: planning, preparation, and execution. The required execution time varies depending on the specific task. Thus, pre-movement brainwaves can serve as a stable basis for predicting actions. This study detects brainwaves covering the entire execution process and analyzes pre-movement brainwave changes to train a deep learning model. This model predicts actions based on brainwave data, overcoming the time delay between brain signal transmission and action execution, enabling faster and more stable real-time control.
This study employs a six-channel LSTM (Long Short-Term Memory) network combined with a Multi-head Attention mechanism. The LSTM encodes brainwave sequence data, retaining temporal information, while the Attention mechanism dynamically weights critical parts of the sequence, enhancing the model′s ability to identify movement intentions. The experimental design includes four movement states (left hand, right hand, both hands pressing keys to move forward, and rest). Brainwave data generated by participants in these states served as training data. The system comprises offline and online phases: in the offline phase, participants performed key-press operations to generate pre-movement brainwave signals for model training; in the online phase, a sliding window outputs the model′s results every 0.1 seconds, and a voting mechanism is applied during the decision phase to enable participants to control a virtual environment character′s movements using real-time brainwave signals.
The experimental results show that the offline models achieved an average accuracy of 86.3% across six participants. ERD/ERS analysis verified that observing movements effectively activates brain regions associated with motor activity, enhancing the state of movement preparation. Moreover, compared to traditional electromyography (EMG) systems, the brainwave-based control system demonstrated lower latency and higher real-time performance. This study provides an innovative brainwave training system with potential applications in neurorehabilitation, virtual reality interaction, and brain-computer interface technologies. Ultimately, the goal is to enable behavior control purely through brainwave signals, without requiring actual physical movement. | en_US |