Transfer Learning System for Attention Network Task EEG signal

The EEG classifier is considered as a critical component for brain-computer interface task systems. There are two traditional challenges for creating these kinds of classifiers. Acquiring and capturing EEG data is a very difficult task and feature selection and extraction are very time -consuming. A new model was designed to overcome these challenges, based on the existing deep learning model. A novel Attention Network Task Dataset is used for this task. To do this, Attention Network Task was conducted and the brain signal was captured during the task by BCI system. The data are preprocessed and transformed by wavelet transformed to the image. This study suggests that transfer learning is a promising method for BCI classification systems. In order to implement transfer learning, Densenet pre-train model is used and the accuracy of the classifier is 80%. Robustness of the model are among the advantage of this new framework.

Keywords:

BCI , EEG classification , wavelet , transfer learning , attention network task.