A Calibration-Efficient 64-Target c-VEP Brain–Computer Interface with Grouped Quinary Relative-Color Modulation and TDCA–MLP Decoding

Abstract

Aims: Conventional code-modulated visual evoked potential (c-VEP) brain–computer interface (BCI) systems commonly employ high-contrast binary black-and-white flickering stimuli, which can induce visual fatigue, while time-shift encoding based on a single pseudorandom sequence limits the number of recognisable targets. This study aims to construct a calibration-efficient, 64-target c-VEP BCI system based on quinary relative-colour modulation to improve its multi-target recognition performance. Study Design: A grouped modulation strategy was applied to a quinary relative-colour stimulus design (red, green, blue, yellow, and grey) for the first time, and a comparative experiment involving 10 participants was conducted to compare the classification performance of conventional decoding algorithms and the proposed hybrid algorithm. Place and Duration of Study: Tianjin Information Sensing and Intelligent Control Key Laboratory, Tianjin University of Technology and Education, Tianjin, China, between April 2026 and June 2026. Methodology: This study proposes a hybrid decoding algorithm based on task-discriminant component analysis and a multilayer perceptron (TDCA–MLP). The algorithm employs TDCA for spatial filtering and dimensionality reduction and a lightweight MLP to model nonlinear features for 64-target classification. Results: The mean classification accuracies across the 10 participants were 85.72% ± 8.52%, 87.56% ± 7.00%, 85.78% ± 8.55%, 91.27% ± 5.06%, and 93.72% ± 3.31% for CCA, FBCCA, TRCA, TDCA, and TDCA–MLP, respectively. A Friedman test revealed a significant overall difference among the five algorithms ( ,P < .001, Kendall’s W = 0.601). Post hoc two-sided Wilcoxon signed-rank tests with Holm correction showed that TDCA–MLP significantly outperformed CCA and FBCCA (adjusted P = .020) as well as TRCA and TDCA (adjusted P = .031). Conclusion: The calibration-efficient 64-target quinary c-VEP system combined with TDCA–MLP achieved high multi-target recognition performance with reduced calibration requirements and favourable subjective visual experience.