EEG Classification Based On Knowledge Accumulation

Brain-computer Interface is a new kind of communication method. It establishes a direct communication between human brain and external environment without the help of neuromuscular pathways. So that, it can not only help some disabled people to regain the ability of communicate with the outside world, but also to provide a special control method for certain areas. In addition, it can provide a new method to explore the mysteries of human brain by making deep study about the brain-computer interface.And there are many difficulty and important problem should be resolved. The study of classifying EEG signals is one of them.When classify an EEG signal, the first thing which the conventional methods will take is that to find the most distinguishing part of the signal. Then classify the signal. Even though, these methods may have a good performance in this time segment which was confirmed by statistical analysis. But they cannot always get a good result in practical applications. Because these methods lost many useful message that were involved in this segment before the selected segment. So that, a new method which was based on knowledge is proposed in this paper. This new method make two ways to overcome the drawback of traditional methods. First, using different windows to intercept signal.These windows have the same starting point, but the length of them are successively longer along the time direction. With the help of these windows the proposed method will take the correlation message between current signal and history data when analysis the current signal segment. After that, make the classification of each window. The result of these windows were called classification knowledge in this paper. Second,when make the final judgment of the current signal the proposed method will combine the classification message which was calculated in each window. In this way, it will accumulate the history knowledge about classifying the signal from each window, and correct the judgment result of these windows to help making a better classification judgment.Before extracting features, the common spatial pattern(CSP) method was used to make per-processing for the original signal. That is a useful method about EEG signal analysis and is widely used in the study of brain-computer interface. But the EEG data which is processed by CSP should be collected from multi-channel. And if the channel number of EEG data is too small then the performance of CSP method will be limited badly. To address this issue this paper combine the harmonic wavelet packet decomposition method with CSP to analysis EEG data. The hybrid approach use the discrimination ability harmonic wavelet packet decomposition method to decompose the EEG data of every channel and get many sub signals in different frequency banks. This discrimination method like the binary wavelet packet decomposition which could decompose the data to different layers, but it could directly decompose the signal to the specified layer. So that, we use the harmonic wavelet packet decomposition method to decompose the signal first. And handling the decomposed signal by CSP method. This method can not only overcome the drawback of CSP when analysis the less-channel data, but also achieve the use of frequency message by decompose the signal in different frequency bands.At last, in order to prove the effectiveness of proposed method in this paper, we use two national BCI competition database to test this method. And compared it with other state-of-the art approaches. The results show that the proposed method could fully use the information of signal and get a high performance and robustness.