In this paper, we propose a kernel canonical correlation analysis (KCCA) based idle-state detection method for asynchronous steady-state visual evoked potential (SSVEP)-based brain-computer interface (BCI) systems. KCCA method can offer a flexible nonlinear solution to adequately extract nonlinear features of multi-electrode electroencephalogram signals. Based on this method, an ensemble KCCA coefficients feature model is proposed by weighting effectively multi-harmonic information and afterwards a threshold classification strategy for idle-state detection is presented. The weights of the model and optimal threshold are trained by the presented parameters learning scheme. Using our method, offline analysis was performed on 10 subjects with 8 fixed common electrodes. The results showed that the idle state could be detected with 95.9% average accuracy when SSVEP could be determined with 93.8% average accuracy. Further, the analysis verified the effectiveness and significant superiority of our method over other widely used ones.
