NPD-RSVP：一种用于目标识别的新型脑机接口范式和解码方法

Rapid serial visual presentation (RSVP) based brain-computer interfaces (BCIs) can detect and recognize target and non-target objects. In this study, we proposed a novel dual-RSVP paradigm known as negative photographic dual-RSVP (NPD), in which one sequence contains the original target images and the other comprises the corresponding negative photographic images. Compared to the conventional RSVP, the proposed method can provide complementary information about target-related brain signals and efficiently avoid repetition blindness (RB). Based on this paradigm, we further proposed a novel electroencephalography (EEG) decoding method, known as cross-frequency decoupling model (CFDM). To model the periodic temporal changes, we first transformed the one-dimensional data vector of each EEG channel into a two-dimensional data matrix by adopting a period corresponding to the dominant neural oscillation components. To extract more discriminative features, we input two target feature maps into different channels and extracted the spatio-temporal dynamics of different brain regions using depth-wise spatio-temporal convolution kernels of different scales. Experiments were conducted on an expanded dataset of 20 subjects using a rigorous 8:2 train-test split to evaluate the performance of the proposed paradigm. The experimental results show that the proposed method achieved a mean classification accuracy of 94.24% and a True Positive Rate (TPR) of 90.62%. These results jointly demonstrate the effectiveness and advantages of the proposed NPD paradigm and the CFDM model for solving the dual-RSVP recognition problem.