Scalogram sets based motor imagery EEG classification using modified vision transformer: A comparative study on scalogram sets

Abstract

Nowadays, motor imagery (MI) electroencephalogram (EEG) is mainly utilized for brain computer interface (BCI) based prosthetic device developments and involves the accurate classification of EEG signals. However, the major challenges are inter-subject and intra-subject variability, presence of noise and artifacts in the acquired EEG signal, this results in low average classification accuracy. To address these issues, the present work proposes an innovative algorithm based on scalogram set formation and modified vision transformer (MViT) model for classification of EEG data. The proposed scalogram sets formed by organizing scalograms of fundamental wavelets as well as their combinations and the proposed Modified Vision Transformer model employs both serial and parallel feeding of initial patches through consecutive transformer blocks, enhancing information flow and extracting diverse features. To verify and validate the proposed methodology, the BCI Competition IV 2b dataset was utilized. The MViT with Morlet and Shannon scalogram set performed accuracies of 86.34 % for intra-subject and 76.19 % for inter-subject classification. The proposed approach performed best among state-of-the-art methods with an average improvement of 3.46 % for intra-subject and 1.75 % for inter-subject in accuracy highlighting the robustness and reliability of the proposed methodology. © 2025 Elsevier Ltd