Loading...
Please wait, while we are loading the content...
Similar Documents
EEG-based BCI Cursor Control:A Classification Model With Convolutional Neural Networks
| Content Provider | Semantic Scholar |
|---|---|
| Author | Jing, Wu |
| Copyright Year | 2018 |
| Abstract | Brain-computer interface(BCI) systems are providing humans with the ability to control external devices like computer cursor using only their thoughts. While invasive BCI systems acquire neural signals with intracranial or subdural electrodes which have the inherent risk of surgery and gradual degradation of signal integrity, noninvasive BCI systems typically acquire neural signals with scalp electroencephalography(EEG)[1]. A challenge for modeling cursor movements from EEG data is to find the representation that could preserve the inherent properties of EEG data and is simple to handle and review[2]. A novel approach for learning the representations from multi-channel EEG time-series proposed by Bashivan et al in 2016[3]draws my interest. In traditional EEG data analysis, the spatial information of the electrodes is usually ignored. While in this novel approach, EEG activities are transformed into a sequence of topology-preserving multi-spectral images, which preserves the inherent spatial and spectral properties of EEG signals. With these featured images, we proceed to build a binary classifier between horizontal cursor movement and vertical cursor movement with Convolutional Neural Network. It is assumed that in the real-time cursor control application, the classifier could be employed to indicate whether the subject wants to move horizontally or vertically. With the orientation settled, we could pass the EEG signal to the regression model to predict the velocity. My proposed binary classifier yields a satisfying accuracy rate at 79.17% with an acceptable training time. |
| File Format | PDF HTM / HTML |
| Alternate Webpage(s) | http://www.jics.utk.edu/files/images/recsem-reu/2018/bci/Report-A.pdf |
| Language | English |
| Access Restriction | Open |
| Content Type | Text |
| Resource Type | Article |
