Abstract:In view of feature extraction of motor imagery electroencephalography (EEG) in the study of brain-computer interface (BCI), a novel method is proposed based on adjacent matrix decomposition of functional brain network in this paper. First, multi-channel EEG signals were used to construct the functional brain network (FBN). Secondly, the singular value characteristic vector was obtained by means of singular value decomposition on the corresponding adjacent matrix, and then the characteristic parameters of EEG were defined based on it. Finally, the parameters were used as input vector for support vector machine (SVM)，and then six datasets of BCI Competition IV Data Sets 1 were used for classification. Experimental results show that different motor imagery tasks are successfully identified by the method of SVM classifier combined with feature extraction which is based on adjacent matrix decomposition of FBN. This method provides a new idea for the study of extracting EEG features.