Research On Recognition System Of Multi Channel EEG Based On Specific Mental Tasks

BCI (Brain Computer Interface) provided a new concept of communication and control method for the handicappeds who can't recover through traditional medicinal technique. Using several different mental tasks'conversions to change control mode was one of implements of BCI. Beside the data acquisition device, this system requires many specific mental tasks and advanced signal processing methods. This paper provides a pattern recognition system of multichannel brain electrical signal. It trains offline, discriminates online, can distinguish several mental tasks such as complex multiply at rate above 70%.Firstly, a medical equipment for detecting encephalopathy was used to collect brain electrical activity. A program was designed to control the device and acquires data after successfully calling the device's DLL in Matlab.The 10%~20% international standard electrodes system contains 16 electrodes'position. Sometimes they are fully redundant so we want to reduce them. After discussing the power spectral intensity and features'separability between the electrodes, the channels of 5~10 were decided finally.There must be too much noise in the brain electric signal. This paper presented two methods, wavelet decomposition-reconstruction and independent component analysis to remove it.Mental tasks were critical in the system. Plenty experiments help to decide 5 applicable ones such as complex multiply, letter composing, virtual count etc. Necessary analyses about divisibility were also done to judge them.4 different types'features were generated, namely AR model parameters, power spectral frequency band intensity, energy for wavelet packet decomposition, wavelet packet entropy. Every type of features were extracted respectively using PCA and ICA method and classified using linear neural network, KNN and BP network. After compared these methods, power spectral frequency band intensity, PCA and linear network were choose to carry out the recognition system.Finally, this paper provided software of GUI as well as a group of simulink blocks to operate data and acquire results. And some elementary experiments...