The current brain-computer interfaces (BCIs) based on steady-state visual evoked potential (SSVEP) typically employ single recognition algorithms, which often result in low accuracy for different time durations. A novel SSVEP recognition algorithm was proposed, the proposed algorithm combined filter bank canonical correlation analysis (FBCCA) and power spectral density (PSD) analysis, in order to improve the universality and accuracy of SSVEP recognition. The proposed method utilized FBCCA to identify highly similar reference frequency signals and then locked in the final response frequency through multiple set of PSD analysis, achieving frequency recognition without the need for training. Experimental results demonstrate that with the stimulation duration of 1 s, the proposed method achieves 86.61% accuracy, 5.44% better than the PSD analysis method, 10.38% better than the canonical correlation analysis (CCA), and an 8.86% better than the FBCCA.