Electroencephalography Signal Analysis and Classification Based on Deep Learning

Brain computer interface (BCI) bridges the interaction between the brain activities and an external machines/device. Electroencephalography (EEG) has its advantages over other brainwave monitoring tools in cost, portability and monitoring frequency and accuracy. The interpretation of the relatively large amount of EEG signals is of key importance to understand brain functionality. Traditional machine learning and signal preprocessing methods alone fail to provide robust and in-time EEG signal interpretation and partially relied on professionally trained expert. For further developments and applications of robust and in-time EEG interpretation, we investigated the application of the deep learning models on a classification task of a motor imagery EEG signal dataset with both spatial and temporal information. We adopt sliding window with specified window sizes and strides to generate training samples for deep learning models and standardize the training samples to improve the model performance. Convolutional neural network, variants of recurrent neural network are investigated and compared on the classification and interpretation performance over the interested dataset. The convolutional neural network demonstrates superior performance in terms of both training efficiency and accuracy to other models.