The Research Of Technology Of Multi-Pattern Recognition On Signal Of Brain-Computer Interface

EEG is one of the body’s vital physiological signals as a weak signal produced by the human brain. The human brain is the most mysterious and most complex organ and has long been the focus of research scientists. EEG signals as a direct result of the human brain is the focal point where scientists study. EEG not only contains the human brain daily activities basic signals, but also contains the human brain thinking signal according to spontaneous imagination or external stimuli generated. EEG plays a very important role regardless of the various regions of the human brain development and other detailed features of artificial intelligence research or related field, have a very important role. Brain–computer interface(BCI) is to establish a direct channel of information exchange between the human brain and computers, wheelchairs or other peripheral electronic equipment. The BCI could both simple control computers or other peripheral electronic devices based on the analysis results of thinking human brain EEG signals produced contained in the external environment to achieve the simple control effect. And according to the motion peripheral electronic devices, corresponding to the human brain the external stimuli to generate new thinking EEG, to regain control of peripheral electronic devices purposes. With the increase of the number of patients with communication barriers or the patients with amyotrophic sclerosis, spinal cord injury or other brain diseases, such brain-machine interface technology is also increasing. How to improve the accuracy of analysis results in brain-computer interface technology has become plagued researchers EEG problem.P300 EEG are specific induced electrical activity produced by the nervous system in the body when accept a particular mode of visual stimulation. P300 EEG becomes the commonly used signals because it has generated a specific time in a specific region of the brain scalp obvious energy distribution, and other characteristics can easily be detected. The experimenter without special training can achieve the desired effect when using BCI system which consists of P300 EEG. The data of this topic comes from dataset II of BCI competition III challenge, conduct research to identify multi-mode Fusion BCI signals in order to improve the BCI system in EEG analysis and calculation of results accuracy, and promote the practical application of brain-machine interface technology.The main contents were involved in this paper as following:1 The pretreatment of P300 EEG. Firstly, The P300 EEG signals should be low-pass filtered because it’s mainly useful is in the 0-30 Hz frequency range. Secondly, the filtered EEG signals should be optimal weighted average overlap processed because the processing of EEG acquisition is easily Interaction by adjacent electrodes and the fluctuation which should be on time will delay.2 The P300 EEG after pretreatment should use four most widely used algorithms like Wavelet transform, AR model, principal component analysis and approximate entropy to extract features. Wavelet decomposition mainly extracts the potential signal from the 2-8Hz band as a test sample. AR model algorithm mainly constructed a 4-order model, selecting the model coefficients for the desired characteristics; Principal component analysis of 64 channels spatially decoupled and the first ten principal components extracting channel as a feature. Approximate entropy algorithm is approximate entropy signal segments selected 10 channels can be calculated as a feature.3 Four kinds of classifier like Support vector machine, BP neural network, RBF neural network and naive-Bayes are used to classification the features of P300 EEG. Support vector machine classifier selects the highest classification accuracy as a linear function of the classifier kernel function. BP neural network classifier constructed layers BP neural network and choice a S-shaped function as transfer function.RBF neural network classifier can approximate any nonlinear function and have good generalization ability and learning speed. Naive Bayes classifier is a simple probabilistic classifier based on the assumption of independence Bayes’ theorem.4 Four kinds of feature extraction algorithm and feature classification algorithms constitute a total of 16 cross-integration of EEG data processing model. Use this 16 kinds of model for P300 EEG test and get different treatment accuracy. After some group of P300 EEG test, the data processing model with Wavelet transform algorithm and Support vector machine classifier get the highest accuracy rate, an average of 90%.