Recognizing EEG signals for brain-computer interface based on machine learning
A brain-computer interface (BCI) basically gives a second chance to people with motor disabilities to be able to communicate again by translating the brain activities (EEGs) into machine-learning languages which in turn controls the devices. However, EEGs are non-stationery rhythms with low amplitud...
Saved in:
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Final Year Project |
| Language: | English |
| Published: |
2019
|
| Subjects: | |
| Online Access: | http://hdl.handle.net/10356/78286 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Institution: | Nanyang Technological University |
| Language: | English |
| Summary: | A brain-computer interface (BCI) basically gives a second chance to people with motor disabilities to be able to communicate again by translating the brain activities (EEGs) into machine-learning languages which in turn controls the devices. However, EEGs are non-stationery rhythms with low amplitudes and high signal-to-noise (SNR) ratio. In order to recognize the motor function related information embedded in the rhythms, a machine learning method was introduced in order to extract the wanted features.
This project describes a generalized common spatial patterns (CSPs) for extracting EEG features while comparing the performance between CSP Feature Selection (FS) and CSP Feature Weighting and Regularization (FWR) as the first task. Afterwards, CSP-based extension (RCSP) is introduced in which a prior coefficient was used in the algorithm. EEG data from Berlin BCI Competition III, data set IVa as well as data set IIa from BCI Competition IV are used for training and testing the algorithms. Results obtained from all 4 tests are compared and analyzed to determine the performance of these feature extraction methods.
The experimental results show that FWR outperformed FS in many of the tasks while RCSP results shows better performance in certain conditions as compared to CSP. |
|---|