Using the following system architecture we managed to classify 6 different hand movements (Forward, Backward, Left, Right, UP, Dawn) for both the real movement and the imagery movement (Motor imagery)
-
Paper can be found at https://academic.oup.com/gigascience/article/9/10/giaa098/5918864#208064346 and the dataset can be found at:https://ftp.cngb.org/pub/gigadb/pub/10.5524/100001_101000/100788/
-
system architecture:
-
Methodology :
we used 3 different feature extraction techniques
- Discrete wavelet transform (DWT)
- Common Spatial Pattern (CSP)
- Extracting statistical features
-
we used different machine learning methods such as LDA, Random forest, and SVM. Also, we used deep learning methods such as CNN, LSTM, and EEGNET architecture.
EEGNET Architecture :
-
Results
-
The following Bar chart shows our results when classifying 6 movements for different subjects (
blue) in comparison with the original Paper (orange)
-
The following line chart shows how the performance change when changing the number of classes for cross-subject results
-
-
Deployment
- we used Unity and Blender for creating a simulator for the movements.
- The simulator can be found here: https://drive.google.com/file/d/1mrWMTimOA4dxspdH9cZ9dznV2-6pDoZW/view?usp=sharing
