Feature Extraction and Classification for Electro-Encephalography Based Bci: A Review

Abstract

A Brain Computer Interface (BCI) system provides a method for controlling a peripheral device. A BCI may use the magnetic, electrical, or metabolic activity of the brain. Electro-encephalography (EEG) is a non-invasive technique. It is popular for BCI research and is preferred due to its high temporal resolution, low cost of devices, convenience and movability. BCI based applications have massive potential in assistive devices, health care, and amusement industry. A regular BCI system comprises of these steps: signal acquisition, pre-processing, feature extraction and classification. An EEG contains the impulsive electrical activity of the brain taken from electrodes placed on the scalp of the subject. The EEG signal is then processed to remove noise and enhance the signal for analysing further. Features are mined from the amplitude and frequency of the recorded analog signals which can be transformed into feature vectors, and given as input to a classifier. Since EEG is non-stationary in nature, vulnerable to artifacts and has high variability, we need algorithms that efficiently extract relevant features and classify the signals accurately. This study reviews some recent applications of BCI and the feature extraction techniques used by them. Machine learning algorithms typically used in EEG-based BCI applications are also studied.

English and Romanian Brain-to-Text Brain-Computer Interface Word Prediction System

“…Feature extraction of EEG signals is one of the important aspects related to the system of BCI due to its important task of obtaining the correct performance of the classification stage [15] because of the complex processes inside the brain [8]. EEG signals are non-stationary, while their spectrum varies with time.…”

“…The algorithms of feature extraction must handle the signal source, which is usually complex and noisy, and detect interesting features [8]. These algorithms are used to detect the features that are strongly connected with the intent of the subject.…”

“…Feature extraction depends on temporal-spatial analysis and/or time-frequency. Some of the techniques used are wavelet transform, autoregressive (AR) models, Fourier transforms (FT), and statistical properties of the signal [8].…”

“…The next step is to translate the extracted features into commands to control an external device or to do different mental tasks [8].…”

“…Some features of EEG may have great elective power to observe different patterns of EEG. To design and implement a well-trained model, some features with high discriminating ability are required [8].…”