Multi-scale Architecture for High Performance RSVP Classification

The ability to directly control external equipment using electrical impulses to translate brain activity into commands or information has made brain-computer interfaces (BCI) an increasingly relevant research topic. Numerous uses for this technology have been discovered, including cursor control, wheelchair navigation, and use in diverse sectors. In this paper, a unique EEGNet-inspired Multi-Scale architecture is proposed, which is intended to boost Rapid Serial Visual Presentation (RSVP) task categorization performance. The performance of our suggested architecture is assessed using the datasets from BCI Competitions II and III, and accuracy of 76.3%, 84.2%, and 90.5% is achieved for subjects II-A, II-B, and IIb, respectively. The average symbols recognized under repetition metric is also used to evaluate the proposed architecture. These results show a notable improvement in accuracy and other evaluation metrics compared to previous research. Overall, the proposed architecture advances the field of BCI technology and offers a fresh, potentially effective method for enhancing RSVP categorization task performance.