Deep Motor Feature in EEG signal processing, for brain computer interface (BCI)

Background and Aim: In this paper we present multi-scale deep convolutional neural networks to work in Electroencephalography (EEG) signals for imagined motor. We have some suggestion for the learning of a set of high-level feature representations by deep learning which is purposed as Deep Motor Features (DeepMF), that is used in brain computer interface (BCI) with imagined motor tasks. Since the extracted DeepMF is different for various tasks and similar for the same tasks. So it is convenient to separate different EEG signals for imagined motor tasks apart. Our approach achieves ۱۰۰% accuracy for ۴ classes imagined motor EEG signals classification on Project BCI EEG motor activity dataset. Moreover, thanks to the highly abstract features DeepMF learned, only ۴.۱۲۵ seconds trials of training data are needed. This DeepMF is compared with the conventional BLDA (Bayesian Linear Discriminant Analysis) algorithm for ۸.۷۵ seconds trials to achieve the same accuracy. Accordingly the BCI response time and the required trials for training are almost declined by half. The experiments are provided to illustrate the effectiveness of the proposed design approach. Methods: deep learning Results: Accordingly the BCI response time and the required trials for training are almost declined by half. The experiments are provided to illustrate the effectiveness of the proposed design approach. Conclusion In this paper, the effective high-level features of EEG representation for brain computer interface based on the multiscale deep convolutional neural networks are proposed. The features extracted from EEG signals boost the performance on the accuracy of imagined motor tasks identification into ۱۰۰%, while only requiring ۴.۱۲۵ seconds of training trials compared with the demand of conventional BLDA for ۸.۷۵ seconds to achieve the same accuracy.

کلیدواژه ها:

EEG ، convolutional ، CNN ، BCI ، deep learning