Multiclass Motor Imagery EEG Signal Classification using FBCSP-CNN

• Published in: Journal of information systems engineering & management Vol. 10; no. 4; pp. 1031 - 1039
• Main Author: Rajesh R. Bhambare
• Format: Journal Article
• Language: English
• Published: 30.04.2025
• ISSN: 2468-4376; 2468-4376
• DOI: 10.52783/jisem.v10i4.10194

Recently, brain-computer interfaces (BCIs) have gained more attention. One of the BCI tasks is to categorise EEG motor imagery (MI). A significant amount of work has been put into MI categorization. Even with significant breakthroughs, multimodal MI decoding remains unsatisfactory. The spatiotemporal-frequency properties of distinct MI were retrieved using a filter bank common spatial pattern (FBCSP) method. The MI EEG recognition model's performance is substantially influenced by the EEG's operational frequency band. Unfortunately, because they employed a vast frequency range, the majority of algorithms failed to successfully harness discrimination from many sub-bands. Consequently, extraction of discriminative features from EEG signals using convolutional neural networks (CNNs) with different frequency components might be a promising method for multi-subject EEG identification. In order to verify this suggested method, experiments were conducted using the BCI Competition III dataset III a, which is publicly available. The performance obtained (accuracy and kappa) was comparable to the best strategy across the comparisons. The findings showed that the FBCSP-CNN technique lowered computational complexity while maintaining a mean categorization accuracy of more than 84.52% and an average kappa value of 0.8. Thus, we can say this algorithm is suitable for MI-BCI systems with less data