Abstract
Based on the Fast Fourier Transform(FFT) and Support Vector Machine(SVM) joint methods, this paper proposes a novel EEG signal processing joint method in the complex scene for more than five kinds of EEG signals. This new method improves the accuracy and comprehensive efficiency of EEG signal processing.
First of all, the new joint method adopts normalization for data preprocessing; next, combines FFT and PCA method for feature extraction; finally, using the weighted KNN method for classification. This method is apply to the classification of EEG signals generated when the subjects observe 0-9 digits. The result demonstrates that accuracy and comprehensive efficiency of the novel joint method are 84% and 87% respectively. It indicates that the proposed joint method can be apply in the complex scene for multi-class EEG signal processing.
Introduction
Electroencephalography (EEG) signal is an electrical signal generated by spontaneous electrical activity of hundreds of millions of neurons in the brain. It contains abundant information about brain activity. EEG is an intuitive reflection of the state and function of the brain, and has the characteristics of complexity, time varying and volatility [1]. At the same time, the brain is also an important organ of human body, which has important research value. In order to facilitate the application of EEG signal in clinical and in practice, researchers have proposed the non-invasive Brain-Computer Interface (BCI) technology [2].
The EEG-based non-invasive BCI technology is a novel human-computer interaction method. It is independent of human peripheral nerve and muscle tissue, and it can realize the communication and control between human brain and computer or other peripheral devices. After the researchers extracted the signal characteristics based on the acquired EEG signals, use machine-learning methods for classification and identification [3].
The ultimate goal is to control the external device or issue commands to the control program. At present, time-domain and frequency-domain analysis methods, such as Fast Fourier Transform (FFT), are the general method of EEG signal processing. Then, machine learning methods, such as Support Vector Machine (SVM) and artificial neural network, are used to classify EEG signals [4].
There are many researches about the classification of EEG data. Through collecting EEG signals from observing different types of images, GIANNAKAKI et al. [3] obtained three types of EEG signals representing calm, positive and negative emotion respectively. The researchers used wavelet transform combined with SVM method to classify the signal.
The accuracy of the result reached 75.12%. YILDIZ et al. [4] studied the classification of EEG signals of epileptic patients and normal people. He extracted features from time domain, frequency domain and power spectrum, and used machine learning methods such as SVM and multilayer perceptron neural network to classify EEG. The accuracy was over 90%. Other researchers have also obtained high classification accuracy in the study of EEG classification.
At present, the research on EEG classification is mostly focus on the simple scene of two-kind or three-kind, but there are few researches on the classification of more kinds of EEG signals. Now there are dataset of 10-kind EEG, but in the known literature, there are very few to deal with such data, so this paper studies classification on this kind of EEG data. So far, there are two widely used joint processing methods for EEG signals in academic circles: FFT-based joint method [5] and SVM-based joint method [6]. However, there is stillroom for improvement in accuracy, time consumption and overall efficiency of these two existing.
