Search

Motor imagery (MI) based brain-computer interface (BCI) is a research hotspot and has attracted lots of attention. Within this research topic, multiple MI classification is a challenge due to the difficulties caused by time-varying spatial features across different individuals. To deal with this challenge, we tried to fuse brain functional connectivity (BFC) and one-versus-the-rest filter-bank common spatial pattern (OVR-FBCSP) to improve the robustness of classification. The BFC features were extracted by phase locking value (PLV), representing the brain inter-regional interactions relevant to the MI, whilst the OVR-FBCSP is used to extract the spatial-frequency features related to the MI. These diverse features were then fed into a multi-kernel relevance vector machine (MK-RVM). The dataset with three motor imagery tasks (left hand MI, right hand MI, and feet MI) was used to assess the proposed method. Experimental results not only showed that the cascade structure of diverse feature fusion and MK-RVM achieved satisfactory classification performance (average accuracy: 83.81%, average kappa: 0.76), but also demonstrated that BFC plays a supplementary role in the MI classification. Moreover, the proposed method has a potential to be integrated into multiple MI online detection owing to the advantage of strong time-efficiency of RVM.