A Deep Learning Method for Classification in Brain-Computer Interface

Abstract

Neural activity is the controlling signal used in enabling BCI to have direct communication with a computer. An array of EEG signals aid in the selection of the neural signal. The feature extractors and classifiers have a specific pattern of EEG control for a given BCI protocol, which is tailor-made and limited to that specific signal. Although a single protocol is applied in the deep neural networks used in EEG-based brain-computer interfaces, which are being used in the feature extraction and classification of speech recognition and computer vision, it is unclear how these architectures find themselves generalized in other area and prototypes. The deep learning approach used in transferring knowledge acquired from the source tasks to the target tasks is called transfer learning. Conventional machine learning algorithms have been surpassed by deep neural networks while solving problems concerning the real world. However, the best deep neural networks were identified by considering the knowledge of the problem domain. A significant amount of time and computational resources have to be spent to validate this approach. This work presents a deep learning neural network architecture based on Visual Geometry Group Network (VGGNet), Residual Network (ResNet), and inception network methods. Experimental results show that the proposed method achieves better performance than other methods.