Task-related component analysis based on time filter and similarity constraint for SSVEP-based BCI

•Similarity-constrained TRCA has obtained better effect through similarity constraint.•The internal factors of local samples are ignored.•Add temporally local weight into the covariance calculation of scTRCA.•Considering the internal relation between sampling points, general noise is suppressed. Task-related component analysis (TRCA) serves as the representative target recognition approach of steady-state visual evoked potential (SSVEP)-based brain-computer interface (BCI). Based on different strategies, a variety of improvement methods have been proposed, among which, similarity-constrained TRCA (scTRCA) has obtained better recognition effect through similarity constraint. However, all the calculations in this method are based on the whole window, and the internal factors of local samples are ignored. In order to solve this problem, this paper proposed to add temporally local weight into the covariance calculation of scTRCA. By considering the internal relation between sampling points, general noise can be further suppressed. Based on two publicly available datasets, the Benchmark and BETA datasets, the performance of the algorithm was evaluated under conditions of varying data length, the number of training experiments, and the number of electrodes. The results showed that the proposed similarity-constrained temporally local TRCA (stTRCA) had better accuracy (average improvement of 5 % − 6 %) and robustness than scTRCA. The maximum ITR of stTRCA was increased by 5 % − 6 % on average compared with scTRCA. Therefore, the proposed algorithm has promising potential for enhancing SSVEP detection.