Abstract:
Aiming at the shortcoming that there exist too many channels in motor imagery (MI)-based brain-computer interface (BCI) systems, this paper proposes a channel selection algorithm based on multi-correlation forward searching (MCFS) algorithm such that the performance of BCI systems can be improved via the optimized the channel set. First, a forward searching algorithm is performed on the channel set via the training set. Meanwhile, the trust values of three correlation algorithms are updated with the classification accuracy of the validating set. Then, according to the above trust values, the high-quality channel set is selected, the common spatial pattern (CSP) algorithm is adopted to obtain the motor imagery related features, and the classification model is trained by means of support vector machine (SVM) with a linear kernel. Finally, the proposed algorithm is implemented on two datasets (BCI competition IV dataset I and BCI competition III datasets IVa), by which the average classification accuracy of 81% and 87% are achieved, respectively. Moreover, compared with three other common channel selection algorithms, the proposed MCFS algorithm obtains the highest average classification accuracy. These results show that the proposed MCFS algorithm has superior performance and provides a technical reference for the application of MI-based BCI system. .