Electroencephalograph-Based Hand Movement Pattern Recognition for Prosthetic Robot Control Using a Combination of Long Short-Term Memory and Stacked Autoencoder Methods

Electroencephalograph (EEG) signals have expanded beyond the medical field into control systems. Improving EEG-based control technology is crucial to enhancing the quality of life for people with disabilities, especially in optimizing prosthetic functions. This research proposes a method to control a prosthetic hand robot using a combination of Long Short-Term Memory (LSTM) and Stacked Autoencoder (SAE) architecture based on EEG signals. Offline tests were conducted by adjusting various parameters on LSTM and SAE, achieving an average accuracy of 99.89% in single-subject training, indicating strong potential in functional hand motion pattern recognition. However, in cross-subject testing-where the model was tested on subjects other than those used in training-the performance significantly declined, with an average accuracy of 33.97%.