Classification of multi-class motor imagery with a novel hierarchical SVM algorithm for brain–computer interfaces

• Published in: Medical & biological engineering & computing Vol. 55; no. 10; pp. 1809 - 1818
• Main Authors: Dong, Enzeng, Li, Changhai, Li, Liting, Du, Shengzhi, Belkacem, Abdelkader Nasreddine, Chen, Chao
• Format: Journal Article
• Language: English
• Published: Berlin/Heidelberg Springer Berlin Heidelberg 01.10.2017; Springer Nature B.V
• Subjects: Algorithms; Biomedical and Life Sciences; Biomedical Engineering and Bioengineering; Biomedicine; Brain; Classification; Competition; Computer Applications; EEG; Electroencephalography; Emulation; Feature extraction; Human Physiology; Image classification; Imaging; Interfaces; Mental task performance; Original Article; Radiology; Support vector machines; Wavelet analysis; Motor imagery; Electroencephalography (EEG); Common spatial pattern; Hierarchical support vector machine (HSVM)
• DOI: 10.1007/s11517-017-1611-4
• ISSN: 0140-0118; 1741-0444

Pattern classification algorithm is the crucial step in developing brain–computer interface (BCI) applications. In this paper, a hierarchical support vector machine (HSVM) algorithm is proposed to address an EEG-based four-class motor imagery classification task. Wavelet packet transform is employed to decompose raw EEG signals. Thereafter, EEG signals with effective frequency sub-bands are grouped and reconstructed. EEG feature vectors are extracted from the reconstructed EEG signals with one versus the rest common spatial patterns (OVR-CSP) and one versus one common spatial patterns (OVO-CSP). Then, a two-layer HSVM algorithm is designed for the classification of these EEG feature vectors, where “OVO” classifiers are used in the first layer and “OVR” in the second layer. A public dataset (BCI Competition IV-II-a)is employed to validate the proposed method. Fivefold cross-validation results demonstrate that the average accuracy of classification in the first layer and the second layer is 67.5 ± 17.7% and 60.3 ± 14.7%, respectively. The average accuracy of the classification is 64.4 ± 16.7% overall. These results show that the proposed method is effective for four-class motor imagery classification.