Loading...

Extraction of Event Related Potentials (ERP) from EEG Signals using Semi-blind Approaches

Jalilpour Monesi, Mohammad | 2017

588 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 50091 (05)
  4. University: Sharif University of Technology
  5. Department: Electrical Engineering
  6. Advisor(s): Hajipour Sardouie, Sepideh
  7. Abstract:
  8. Nowadays, Electroencephalogram (EEG) is the most common method for brain activity measurement. Event Related Potentials (ERP) which are recorded through EEG, have many applications. Detecting ERP signals is an important task since their amplitudes are quite small compared to the background EEG. The usual way to address this problem is to repeat the process of EEG recording several times and use the average signal. Though averaging can be helpful, there is a need for more complicated filtering. Blind source separation methods are frequently used for ERP denoising. These methods don’t use prior information for extracting sources and their use is limited to 2D problems only. To address these problems, we propose three new semi-blind denoising methods which are based on tensor decompositions. Also, a new method similar to Common Spatial Pattern (CSP) method is presented in this study called Common Temopral Pattern (CTP). Then, we propose a new feature extraction method in CSP and CTP structures. Finally, we present our combined feature extraction method called CSP-CTP in order to use both spatial and temporal information simultaneously. To validate our proposed method for feature extraction, we applied it to the character detection of P300 speller on dataset II of the BCI Competition III. Our proposed method reached 98.5%, 90.5% and 73.5% average detection accuracies for both of subjects when 15, 10 and 5 repititions are used for each intensification. Our proposed method has the best performance compared to other methods. It’s worth mentioning that computational complexity of our proposed method is far less than most of other methods. This makes it a valuable method for applications that we need to have less run time
  9. Keywords:
  10. Event Related Potential (ERP) ; Common Spatical Patterns ; Electroencphalogram Signal ; Canonical Polyadic (CP)Decomposition ; Denoising Source Separation (DSS)

 Digital Object List

 Bookmark

No TOC