Motor Imagery Multi-Tasks Classification for BCIs Using the NVIDIA Jetson TX2 Board and the EEGNet Network.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Computer Science

Background:

• Brain-Computer Interfaces (BCIs) offer significant potential for assisting individuals with motor disabilities.
• Existing BCI systems face challenges in portability, processing speed, and data accuracy.

Purpose of the Study:

• To implement an embedded, multi-task classifier for motor imagery using EEGNet on an NVIDIA Jetson TX2.
• To develop and compare channel selection strategies for improved BCI performance.

Main Methods:

• Utilized EEGNet integrated with NVIDIA Jetson TX2 for embedded classification.
• Developed two channel selection strategies: accuracy-based and mutual information-based.
• Implemented a cyclic learning algorithm to optimize model convergence and hardware utilization.
• Employed k-fold cross-validation on the HaLT public benchmark dataset for motor imagery EEG signals.

Main Results:

• Achieved average accuracies of 83.7% for subject-specific classification and 81.3% for task-specific classification.
• Processed each task with an average latency of 48.7 ms.
• Demonstrated the effectiveness of the developed channel selection methods.

Conclusions:

• The proposed framework provides a viable solution for online EEG-BCI systems requiring rapid processing and dependable classification.
• This approach addresses key limitations of current BCI systems, enhancing portability and efficiency.