Camacho-Rosa, Jaime J.

Profile Picture

Filters

20181
Reset filters

Settings

Citations

Publication Search Results

Now showing 1 - 1 of 1
  • Thumbnail Image
    Publication
    Motor imagery classification using single-channel eeg signals for brain computer interfaces
    (2018) Camacho-Rosa, Jaime J.; Manian, Vidya; College of Engineering; Hunt, Shawn D.; Arzuaga, Emmanuel; Department of Electrical and Computer Engineering; Valderrama, Clara I.
    Brain Computer Interface (BCI) systems based on motor imagery are systems designed to communicate between a device and a subject through electroencephalogram (EEG) signals by performing an imaginary task, in this case movement of right and left hand. Traditionally, EEG data acquisition has been done utilizing any number of electrodes from 8 to 32 or 64 placed over the scalp. This results in a large amount of data, consequently entails intensive computational algorithms. With the advance of technology and the drive for data to be more manageable and portable without losing efficiency, a Single-channel BCI system is proposed. There has not been many studies with motor imagery classification of EEG signals using a single channel. This thesis will evaluate the performance of several methods for feature extraction, feature selection and classification algorithm to deal with the varying statistical properties in EEG signals during trials, tasks, recordings and, sessions that often leads to deteriorated BCI performance. The performances will be evaluated with binary motor imagery EEG signals recorded in the laboratory. First, a time-frequency decomposition of the signal is performed and texture descriptors such as correlation, energy, contrast, homogeneity and dissimilarity are calculated from the GLCM matrices for each spectrogram sub-band. Another method for feature extraction is Common Spatial Pattern (CSP) which is used to discriminate between the two binary motor imagery classes. The texture descriptors and CSP transform are used to train the Support Vector Machine (SVM) and Logistic Regression classifier. These methods are tested using 10-fold cross-validation to test performance of the BCI system. The stimuli presentations and feedback from the system are implemented in Python and integrated with OpenViBe to implement BCI system in real-time.