An information fusion scheme based common spatial pattern method for classification of motor imagery tasks. (September 2018)

• Record Type: Journal Article
• Authors: Wang, Jie; Feng, Zuren; Lu, Na; Sun, Lei; Luo, Jing
• Is Part Of: Biomedical signal processing and control. Volume 46(2018)
• Page Start: 10
• Page End: 17
• Publication Date: 2018-09
• Subjects: Brain-computer interface -- Common spatial pattern -- Motor imagery -- Information fusion -- Classification; Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2018.06.008
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlights: We propose an information fusion scheme based common spatial pattern to classify motor imagery tasks. Experiments demonstrate that the multichannel EEG signals can be divided into several symmetrical overlapping sensor groups. A novel soft classification model is proposed based on the average Euclidean distance ratio. The proposed method outperforms other existing competitive approaches, especially for the subjects with poor classification performance. Abstract: Common spatial pattern (CSP) as a feature extraction approach has been successfully applied in the field of motor imagery (MI) tasks classification. The classification performance of CSP deeply depends on the MI related channels and classifiers. However, many existing variants of CSP usually design spatial patterns by removing irrelevant or noisy distorted channels and selecting classifiers manually. In this paper, we propose a novel but simple calculation model termed information fusion scheme based CSP (IFCSP). It employs information fusion technology to take the place of conventional classifiers. Firstly, we divide all channels into several symmetrical sensor groups. Then the average Euclidean distance ratio (EDR) of each sensor group is calculated between different MI tasks following CSP. In the end, information fusion technology is employed to make the utmost of EDRs of all sensor groups to obtain the final result. In this study, the channel division scheme and parameter setting are determined by cross-validation on training data. As such, the proposed method can be customized to yield better classification accuracy. The proposed IFCSP method is validated on the well-known BCI competition IV dataset 2a. Experimental results reveal that the proposed IFCSP method outperforms other existing competitive approaches in the classification of motor imagery tasks.