| Brain-Computer Interface(BCI)refers to a new type of communication system,which is directly established between brain and peripheral devices through thinking without relying on traditional neural pathways.BCI system mainly includes three parts: signal acquisition,signal processing and equipment control,the core of which is the signal processing part.At present,the available EEG signals of BCI system mainly include steady-state visual evoked potential,P300 and motor imagery signals,among which motor imagery signals can be generated spontaneously only by imagery,with simple operation and low workload,and are favored by applied BCI researchers.However,as motor imagery signal is a spontaneous signal,the signal characteristics are very unstable,and the difference between individuals is also large.In addition,the most important problem is that it is limited by the type of input signal pattern,and can only provide fewer output instructions.In order to solve the above problems,this paper designs a BCI system based on motor imagery pattern encoding,and designs classifiers which are suitable for private-use and public-use BCI systems based on motor imagery.The main work of this thesis is as follows:(1)Aiming at the problem of fewer output instructions in the traditional BCI system based on motor imagery,the encoding and decoding links are introduced into the system.The different combinations of motor imagery patterns are compiled into multiple output states through the encoding links,and then a variety of instructions are translated through the decoding links.The output instructions are effectively added,which increases the output instructions from the original 2 to 6 without adding any other additional EEG input patterns.(2)Different classifiers are designed to improve the classification accuracy for two kinds of BCI systems based on motor imagery with different application objects: private-use and public-use.As for the characteristics of private-use BCI system based on motor imagery that the use objects are known and EEG data samples can be collected in advance for training,the autoregressive(AR)model is combined with the traditional SVM classifier,and the time-domain features are replaced by frequency-domain features as input of the classifier and the average classification accuracy is improved by 3.42%.As for the characteristics of public-use BCI system based on motor imagery that the use objects are unknown and no data of the subjects can be obtained in advance,a weighted voting classifier based on Hilbert-Huang Transform is designed in combination with event related synchronization/desynchronization phenomenon and the average classification accuracy is increased by 3.62% compared with traditional time-frequency classification method of wavelet transform.(3)Intelligent wheelchair with BCI system based on motor imagery is designed and built.The whole system includes control system,signal processing system,communication system and signal acquisition system.In addition,the hardware selection and debugging are completed,and the corresponding software programming is completed with the platforms of Keil,Matlab and Qt respectively.After the completion of the system construction,the performance of the system is tested,the average classification accuracy with 86.67%,the instruction decoding accuracy with 77.74% and the Kappa value with 0.74 are obtained,and the practicability of BCI system based on motor imagery is verified. |
