Research on two-class and four-class action recognition based on EEG signals

Area of Science:

• Neuroscience and Biomedical Engineering
• Focus on Brain-Computer Interfaces (BCI) and Rehabilitation Robotics

Background:

• Motor disorders significantly impact quality of life, driving demand for advanced rehabilitation solutions.
• Electroencephalogram (EEG) signal analysis is crucial for developing assistive technologies like lower limb rehabilitation robots and exoskeletons.
• Accurate EEG signal recognition is vital for effective human-machine interaction in neurorehabilitation.

Purpose of the Study:

• To design and evaluate a Convolutional Neural Network-Long Short-Term Memory (CNN-LSTM) model for classifying EEG signals.
• To investigate the model's efficacy in recognizing two-class and four-class motion intentions.
• To analyze time-frequency characteristics and event-related potential (ERP) phenomena, including ERD/ERS, for improved EEG signal processing.

Main Methods:

• Development of a brain-computer interface (BCI) experimental scheme.
• Preprocessing of EEG signals to extract relevant features.
• Implementation of a hybrid CNN-LSTM neural network for EEG signal classification.
• Analysis of time-frequency characteristics and event-related synchronization/desynchronization (ERD/ERS) patterns.

Main Results:

• The proposed CNN-LSTM model demonstrated effective classification of both binary and four-class EEG signals.
• The model achieved higher average accuracy and kappa coefficients compared to two other classification algorithms.
• Analysis revealed significant ERD/ERS characteristics useful for motion intention recognition.

Conclusions:

• The CNN-LSTM model shows significant promise for accurate EEG signal-based motion recognition.
• This approach can enhance the performance of brain-computer interfaces in motor disorder rehabilitation.
• The findings support the use of advanced deep learning models for analyzing complex EEG data.