Motor Movement Detection Using EEG Brain Sygnals and Hybrid Recurrent Convolutional Neural Network
- 0 Collaborators
The goal of the project is to improve current methods of BCI systems for motor movement limb timeseries classification. To achieve this goal hybrid convolutional - CNN RNN with little data preprocessing. We plan to evaluate project's results with real-time classification. ...learn more
Project status: Under Development
Groups
Student Developers for AI
Intel Technologies
MKL,
Intel Python
Overview / Usage
Motor movement detection is one of Brain-Computer-Interface in which EEG signals are used. Proposed method allows to train hybrid RNN CNN neural network and then detect limb movement/imaginery. It can be used as both communication interface and help for impaired people.
Methodology / Approach
I trained neural network implemented with tensorflow keras on EEG Motor Movement/Imagery Dataset from Physionet. 1005 electrode mapping has been used to map 64 electrodes to 2D space for Indipendent Component Analysis and noise removal. Neural Network has been validated with cross parient methodology. Average accurancy for test subset was at about 87 percent. Transfer learning pretraining scenario for global weight start has allowed to increase it by 1 percent.
I have also created average heatmap of regions selected as artifacts (attachment no 1).