Sub-Band Consistency-Driven SSVEP Fusion Control Method

To address the insufficient classification accuracy of steady-state visual evoked potential (SSVEP) brain–computer interfaces (BCIs) under short time windows, this paper proposes a sub-band consistency-driven fusion control method integrating Filter Bank Canonical Correlation Analysis (FBCCA) with deep learning models. The proposed method bridges two decoding paradigms via a sub-band consistency arbitration mechanism: the agreement among three independent FBCCA sub-bands in identifying target frequencies is used to evaluate decision reliability. When sub-bands reach consensus, the FBCCA result is adopted directly; in cases of disagreement, a deep learning model is invoked for secondary verification. Leave-one-subject-out cross-validation on 70 subjects from the public BETA dataset demonstrated that, within a 2-s time window across 40 targets, standalone FBCCA achieved an accuracy of 75.96%±19.44%, while the proposed fusion method (FBCCA + SSVEPformer) reached 91.79%±9.72%, improving by 15.83 percentage points, with approximately 52.3% of trials resolved independently by FBCCA, significantly reducing GPU invocation frequency. In a BCI-controlled drone validation experiment, the method attained a 95.0% completion rate in standard flight tasks. These results indicate that the sub-band consistency arbitration mechanism effectively assesses FBCCA decision reliability and enables cascaded, high-precision EEG signal recognition when combined with deep learning models.