An ICA-based spatial filtering approach to saccadic EOG signal recognition. (May 2018)

• Record Type: Journal Article
• Authors: Lv, Zhao; Wang, Yang; Zhang, Chao; Gao, Xiangping; Wu, Xiaopei
• Is Part Of: Biomedical signal processing and control. Volume 43(2018)
• Page Start: 9
• Page End: 17
• Publication Date: 2018-05
• Subjects: 00-01 -- 99-00; Electrooculogram (EOG) -- Human–computer interaction -- Independent component analysis -- Saccade related independent components; 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.2018.01.003
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlights: A spatial filtering algorithm for saccadic EOG recognition was presented. ICA was applied to analyze mapping model between saccadic "source" and electrodes. Saccadic "source" automatic selection and validity judgment method were designed. A saccade sample optimization strategy was developed. Abstract: To establish a stable electrooculography (EOG)-based communication way for the patients with motor diseases, we proposed a saccadic signal recognition algorithm using independent component analysis (ICA) in this paper. According to the mapping pattern of independent components (ICs)-to-electrode, we designed an optimum ICA-based spatial filter. On this basis, we extracted feature parameters of four types of saccadic signals (i.e., up, down, left, and right) by linearly projecting pre-processed EOG signals to the spatial filter. In order to determine saccade related independent components (SRICs) and improve the recognition accuracy, we also developed an automatic SRICs detection algorithm and sample optimization strategy. Under lab environment, we adopted the support vector model (SVM) as the classifier. The average recognition accuracy of unit saccadic signals achieved 99.0% (before sample optimization) and 99.57% (sample optimized) over 10 participants, which reveals that the proposed algorithm presents an excellent classification performance in saccadic signals recognition.