EUVIMEDEUVIMED entdecken
Crossref

An efficient deep learning-based motor imagery EEG classification framework for human–machine interface and assistive robotic control

Mohammed Imad AL-Amrie, Mehdi Nasri, Zied Othman Ahmed, Zohreh Fotouhi

Journal of Simulation and Analysis of Novel Technologies in Mechanical Engineering · 18.07.2026

DOI: 10.71939/jsme.2026.1246039 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Journal of Simulation and Analysis of Novel Technologies in Mechanical Engineering, erschienen am 18.07.2026. Verfasst von Mohammed Imad AL-Amrie und 3 weiteren Autor:innen.

AbstractSchlagwörterBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Motor Imagery (MI)-based Brain-Computer Interface (BCI) systems play a critical role in advancing assistive robotic control and human–machine interface (HMI) technologies. However, due to spectral overlap and the complex ‫ self-attention temporal dynamics of EEG signals, achieving discriminative representations and accurate classification for real-time control remains a significant challenge. In this study, we propose an Attention-based Dual-Branch Time-Frequency Information Fusion Convolutional Neural Network (ADB-TFIF-CNN), which employs a parallel architecture to simultaneously extract frequency-domain and time-frequency features derived from Short-Time Fourier Transform (STFT) and Continuous Wavelet Transform (CWT). By leveraging an attention mechanism, the model captures both focused and complementary representations of MI-EEG signals. The network enhances its discriminative capability by fusing localized time-frequency representations (spectrograms) and dynamic multiscale time-frequency information (scalograms). Furthermore, optimal feature selection based on Mutual Information reduces redundancy and improves classification accuracy. In the decision-making stage, an ensemble learning framework based on Bagging is utilized, integrating outputs from multiple independent classifiers to enhance classification stability and reduce the impact of redundant features. Experimental evaluation on the BCI Competition IV Dataset 2a demonstrates the effectiveness of the proposed architecture for MI signal classification, achieving an average accuracy of 98.32%. These results indicate the potential of the proposed framework for future assistive BCI applications in robotic and mechatronic systems, although further validation under cross-subject, cross-session, and real-world conditions is required.

Schlagwörter

motor imageryEEG classificationbrain-computer interfacehuman-machine interactionattention mechanismdual-branch convolutional neural networktime-frequency feature fusionshort-time Fourier transformcontinuous wavelet transformspectrogram

Zitieren

APA

AL-Amrie, M. I., Nasri, M., Ahmed, Z. O., & Fotouhi, Z. (2026). An efficient deep learning-based motor imagery EEG classification framework for human–machine interface and assistive robotic control. Journal of Simulation and Analysis of Novel Technologies in Mechanical Engineering. https://doi.org/10.71939/jsme.2026.1246039

Vancouver

AL-Amrie MI, Nasri M, Ahmed ZO, Fotouhi Z. An efficient deep learning-based motor imagery EEG classification framework for human–machine interface and assistive robotic control. Journal of Simulation and Analysis of Novel Technologies in Mechanical Engineering. 2026. doi:10.71939/jsme.2026.1246039

BibTeX
@article{ALAmrie2026,
  author = {AL-Amrie, Mohammed Imad and Nasri, Mehdi and Ahmed, Zied Othman and Fotouhi, Zohreh},
  title = {An efficient deep learning-based motor imagery EEG classification framework for human–machine interface and assistive robotic control},
  journal = {Journal of Simulation and Analysis of Novel Technologies in Mechanical Engineering},
  year = {2026},
  doi = {10.71939/jsme.2026.1246039}
}
Kostenlos starten →