Research On Recognition Algorithms Of SSVEP For Practical Brain-Computer Interface

The brain-computer interface based on SSVEP has become one of the research hotspots in BCI field.But at present,most of the SSVEP recognition algorithms are difficult to provide good classification performance for the SSVEP-BCI systems based on the low-precision acquisition devices.In addition,some devices for recording brain signals with high-precision quality are relatively expensive.These factors limit the practical application range of SSVEP-BCI system.Furthermore,for different experimental paradigms or different subjects,the distribution of their brain signals may be different,so that it is difficult to employ old datesets to train new algorithm models.In view of this,a novel SSVEP recognition algorithm,weighted canonical correlation analysis(WCCA),was proposed,which associates a weight with each reference signal,in order to modify the competitiveness of each reference signal.The results show that the average classification performance of WCCA is superior to CCA and the average improvement performance is 2.8%～7.9%.And the results verify the feasibility of WCCA.In order to verify the WCCA optimization idea can improve the classification performance of the existing SSVEP recognition algorithm,this thesis proposed the filter-bank weighted canonical correlation analysis(FBWCCA)by combining WCCA and FBCCA The results show that the classification performance of FBWCCA is superior to that of FBCCA.And as the time window increases,the performance gap between the FBWCCA and FBCCA becomes more and more remarkable.The results also confirm that WCCA and its derivative algorithms are feasible and robust.Finally,in order to solve the problem that it is difficult to employ old datesets to train new algorithm models,by combining tradaboost and WCCA,a weighted canonical correlation analysis algorithm based on transfer learning(TWCCA)was proposed.The result confirms that the recognition performance of TWCCA is superior to that of WCCA in any time window,and under the same training set conditions,the average accuracy of TWCCA is improved by 3.546%.And the result also verifies that TWCCA can effectively improve the reusable quality of the datasets recorded in the past and reduce the training time required by SSVEP-BCI system.