Feature Selection of Motor Imagery EEG Based on Mutual Information and Cooperative Particle Swarm Optimization

To tackle the issue of inadequate feature selection in motor imagery tasks within brain-computer interfaces, an electroencephalography (EEG) feature selection algorithm that integrates mutual information (MI) and cooperative particle swarm optimization (CPSO) with the common spatial pattern (CSP). Initially, the CSP algorithm is employed to extract spatial features from EEG signals. Subsequently, the extracted spatial feature set is refined using MI and CPSO. The filtered spatial feature subset is then fed into a classification model for recognition purposes. The algorithm's effectiveness was evaluated using the Brain-computer interface (BCI) 2b dataset from an international brain-computer interface competition and the BCI-T dataset of acute stroke patients. The outcomes of the experiments confirm that the proposed algorithm achieves average accuracies of 84.03% and 74.17% on the respective datasets, marking improvements of 11.51% and 13.56% over results obtained without feature screening. These findings confirm the algorithm's capability to effectively extract key EEG signal features and its promising potential in the recognition of motor imagery EEG intentions.