Hand-free Electrical Wheelchair Based on EEG Signal Classification

An electroencephalogram (EEG) is a test that measures electrical activity in the brain using electrodes attached to the scalp. The paper is mainly about how we process our raw EEG data and adopt different types of neural networks to classify the extracted signals. First, the raw data is processed by filtering to remove the noise and then applied with ICA analysis to prevent the influence of the EOG signals. After the data is segmented by ERP segmentation, our two main extracted features are determined to be the superimposed average value of the potential in all epochs segmented and the average power after Time-Frequency Analysis. Three different types of neural networks are chosen to classify EEG signals, Multilayer Perception (MLP), Convolutional Neural Network (CNN), and Long Short-Term Memory (LSTM). From the results, it is shown that LSTM gets a better performance in classification than the other two models.