Hand-Motion Recognition Based On EEG And Eye Auxiliary

EEG(Electroencephalogram) is a biological signal generated by the human brain, which record the electrical activity of brain by the arrangement of electrodes on the scalp or intracranial. Currently, EEG-based brain computer interface (Brain-Computer Interface, BCI) technology has become a research focus, but the success experience of using EEG control the prosthetic is rare, recognition of different types of hand motions directly by the EEG. is very difficult In the research, we found that some parts of the body will have some natural reactions when the hand doing the motions (open, close, left-rotation and right-rotation). And the eye-movement is the most obvious in the process. So using the eye-movements to recognize different hand-motions is discussed in this paper. The main work and innovation are listed as follows:(1) A novel notch method based on Hankel and SVD(Singular Value Decomposition) is proposed. The Hankel matrix is build by the EEG signal, and the signal's 50Hz frequency components can be found by adding a certain amplitude of the sinusoidal signal as the"guide signal". Then the power frequency part is removed and the 50Hz frequency part of the source signal is preserved. This approach is differ from other filtering methods, the 50Hz frequency component is usually filtered out completely by the general filter, or some frequency component around 50Hz will be filtered, Which can lead to a certain distortion of the source signal. But the proposed method maximizes the reduction of the signal, which is more conducive to the later EEG analysis.(2) The EEG acquisition experiment is designed base on the comparative experiments of hand motion imagery with eye closed and hand motion imagery with eye auxiliary. Because of EEG's weak and sensitivity character, to avoid unnecessary interference, the subject is required to self-proceed rather than conduct by any stimulation during the experiments, and the action time is recorded by another assistant. In order to study more comprehensive hand motions EEG patterns, two experiments is designed: The first one is hand motion imagery experiments with eyes closed, the subject is required to imagine right hand to do open, closed, left rotation, right rotation imagery motions, while his right hand do the corresponding action. The Second one, hand motions imagery experiments with eye auxiliary, the subject is required to placed the right hand on the table, and have a 20 ~ 30 cm distance between eye and the hand. The distance can not be changed in the entire process and his eyes should watch the thumb when the motion is action.(3) Three methods for the hand motions EEG recognition are proposed. Option one: wavelet entropy. Each EEG rhythm is decomposed by the wavelet packet, and the wavelet entropy is computed by the corresponding wavelet coefficients. Option two: analyze the EEG based on the dynamic energy of signal recognition. The EEG is wavelet packet decomposed in different C3 rhythm bands to different sections, with 50 sampling points for the step, a step away from the sampling point for the calculation of dynamic EEG rhythms energy as the motion features. Option three: EEG-basedμrhythm (EEG 9-11Hz band) space energy recognition. Using wavelet packet decomposition to getμrhythms to built space energy as the features, and put into the Elman network to identify. According to the recognition results, the performance is not good enough..(4) Research the hand motions EEG under two statuses: eye closed and eye auxiliary. First calculate the EEG wavelet entropy of the two status, the result is proved that the EEG complexity is lower under the eye auxiliary status. And then using the SVD to extract the eye movement information as the features and put into the Elman network to recognize four hand motions.(5) An innovative method using the native eye-movement to recognize the hand motions is proposed. When the previous researchers analyzed the eye-movement of auxiliary control prosthesis, they usually make the eyes do the movements (such as eyeball clockwise rotation, anticlockwise rotation, left shift and right shift) to classify the hand motions. Although they have achieved a high recognition rate, the process look less"nature". In this paper, we use eye-movement information exacted from the EEG to recognize four hand motions, according to the 84% accuracy, this method is proved feasible.