A novel precisely designed compact convolutional EEG classifier for motor imagery classification

• Published in: Signal, image and video processing Vol. 18; no. 4; pp. 3243 - 3254
• Main Authors: Abbasi, Muhammad Ahmed, Abbasi, Hafza Faiza, Aziz, Muhammad Zulkifal, Haider, Waseem, Fan, Zeming, Yu, Xiaojun
• Format: Journal Article
• Language: English
• Published: London Springer London 01.06.2024; Springer Nature B.V
• Subjects: Accuracy; Classification; Classifiers; Computer Imaging; Computer Science; Datasets; Electroencephalography; Human-computer interface; Image classification; Image Processing and Computer Vision; Machine learning; Multimedia Information Systems; Original Paper; Pattern Recognition and Graphics; Principal components analysis; Robustness; Signal,Image and Speech Processing; Spatial data; Vision; Motor imagery (MI); Brain–Computer interface (BCI); Electroencephalography (EEG); Multiscale principal component analysis (MSPCA); Compact Convolutional EEG Classifier (CCEC)
• ISSN: 1863-1703; 1863-1711
• DOI: 10.1007/s11760-023-02986-1

Robust classification of electroencephalogram data for motor imagery recognition is of paramount importance in brain–computer interface (BCI) domain. Since EEG signals are highly subject-dependent, inter-subject variations can greatly impair the robustness of motor imagery (MI) classification. Therefore, this study introduces a precisely designed deep learning architecture namely compact convolutional EEG classifier (CCEC) which achieves better performance in both precision and efficiency. Specifically, the recorded EEG signals are first denoised using multiscale principal component analysis (MSPCA) technique. Then, such raw EEG data are converted into small tempo-spatial data matrices with a two-step signal preprocessing technique. Finally, the tempo-spatial matrices are fed to the proposed CCEC model for MI classification. Experimental results on two benchmark datasets demonstrate that the proposed model not only performs exceptionally well in subject-specific case with an average classification accuracy of 98.2% on dataset 1 but also shows a reasonable average classification accuracy of 72.64% in the subject-independent case. Additionally, with a mere 10% adaptation to subject-specific data, a further improvement of 18% is achieved, thus attaining a noteworthy 90% accuracy in the inter-subject classification. Results also reveal that the proposed CCEC model is highly robust to noisy data, ensuring reliable performance in real-world scenarios.