| Electroencephalogram(EEG)signal is a kind of bioelectric signals associated with human's thinking activity, which is recorded by electrode put into brain or surface of the brain. People's limb movement is commanded by brain and performed by skeletal muscle. This paper gets some inspiration from brain nerve science, which is that people's limb movement has some correlation with the activity of the brain, so trying to control the hand motions based on EEG is a good choice. The control information of prosthetic hand can be got by analyzing the corresponding relation between limb movement and EEG which is recorded when the hands move. Nowadays, the control research of prosthetic extremity by EEG is concentrated on assisted stimulation, the recognition accuracy is high in this method, but it needs a lot of external installations and is complicate in operation. The operation of spontaneous EEG which controls prosthetic hand is easy and practical, but the research in this method is rare because of the low recognition accuracy.From the perspective of clinical application, this paper begins to study on spontaneous EEG assisted by corresponding eye-motion which controls the 4 hand movements (namely, hand opening up and down, hand opening and closing). The main work is as below:(1) As there is obvious difference inαrhythm wave of EEG between the eye opening and eye closing, further analysis indicates that theαrhythm wave relative energy is larger when eye closes than that when eye opens. Because of this, in the way of controlling signal source, the assisted eye-motion is used when the hands move. Hand opening up is assisted by eye moving up, hand opening down is assisted by eye moving down, hand opening is assisted by eye opening, hand closing is assisted by eye closing.(2) In the preprocessing stage, to eliminate the noise mixing in EEG generated by interference source, theμlaw threshold method in the second generation wavelet is selected, because this method has better de-noising effect than other threshold methods in the experiment analysis. To eliminate the disturbance mixing in EEG generated by independent source, which can't be eliminated by wavelet transform's method, Fast ICA Algorithm is used. The results show that theμlaw threshold method in the second generation wavelet has better de-noising effect and Fast ICA Algorithm has more ideal separate performance.(3) In the feature extraction phase, chaos analysis and wavelet packet transform are applied. First, C3,C4,P3 and P4 are selected as the best EEG collecting points when the hands move, then the signals acquired from these channels are preprocessed. After that, the EEG features are extracted by the above two methods, the feature vectors of maximum Lyapunov exponents and correlation dimensions are obtained in the former method; the 4 feature rhythm waves are got in the latter method, the ratios of every rhythm wave energy to the sum of 4 rhythm wave energy are used as the feature vector. Then the 24 dimension feature vectors can be obtained through combining all the feature vectors.(4) In the pattern recognition phase,"one to all"SVM multiple pattern classifier is applied to recognize four movement patterns based on EEG. This way is well used in clustering method ahead of classifying method of binary tree's more classifying algorithm.The experimental results indicate that the assisted eye-motion method can effectively discriminates the four hand-motion patterns with the 83.7% correct rate, which is better than spontaneous EEG recording or spontaneous EEG recording combined with hand movement. The recognition correct rate of combining with chaos analysis and wavelet packet transform is more ideal than the any way of the two. The SVM classifier is better than BP,Elman,RBF neural network in classifying ability. |
PDF Full Text Request