Error Related Potential Detection Algorithm for Brain Computer Interface

The utilization of error-related potentials (ErrPs) in brain-computer interfaces shows great promise for improving control performance. However, several limitations, such as insufficient data and data imbalance, often impede the accurate recognition of ErrPs. This study proposed a Filter Bank Convolutional Neural Network (FBCNN) method to detect ErrPs. The architectural design integrates signals from various frequency bands and employs a Convolutional Neural Network (CNN) to capture spatial and time information within the signals. Furthermore, a time window that randomly shifts is devised to address variances stemming from different response times among subjects. Simulation experiments affirm the effectiveness of the proposed methodology. We also analyze the enhancement in system control performance by using ErrPs in correcting and rejecting erroneous commands in two scenarios.