Artifact removal for emotion recognition using mutual information and Epanechnikov kernel. (May 2023)

• Record Type: Journal Article
• Authors: Grilo, Marcelo; Moraes, Caroline P.A.; Oliveira Coelho, Bruno F.; Massaranduba, Ana Beatriz R.; Fantinato, Denis; Ramos, Rodrigo P.; Neves, Aline
• Is Part Of: Biomedical signal processing and control. Volume 83(2023)
• Publication Date: 2023-05
• Subjects: Blind Source Separation (BSS) -- Artifact removal -- Electroencephalogram (EEG) signals -- Emotion classification -- Epanechnikov kernel; 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.2023.104677
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlights: New method for artifact removal in EEG signals. Mutual Information based algorithm using Epanechnikov kernel is used. The method outperforms classical methods such as SOBI. Gaussian kernel was also tested having a slightly inferior performance. Methods were evaluated in an emotion classification context. Abstract: This work proposes the use of two methods for artifact removal in Eletroencephalogram (EEG) signal analysis. Two new Blind Source Separation (BSS) algorithms are used to this end, based on Mutual Information (MI): one including probability density estimation through the use of Gaussian kernel and another through Epanechnikov kernel. For comparison, the classical SOBI algorithm was used. For performance evaluation, a scenario of emotion recognition through EEG signals was considered. After extraction of the latent independent components by the BSS algorithms, MARA was used for artifacts automated identification and removal. High Order Crossings (HOC) and Hjorth features were extracted from the cleaned EEG signal and used as inputs to a SVM classifier. The results obtained show the importance of artifact removal, since the application of the simplest method is already able to attain a gain of about 12 %. In addition, it is shown how the algorithm based on the Epanechnikov kernel has the best performance, leading to an accuracy of 80.13 %, with the advantage of being simple and presenting a lower computational cost when compared to the algorithm obtained with the Gaussian kernel.