P300 based character recognition using convolutional neural network and support vector machine. (January 2020)

• Record Type: Journal Article
• Authors: Kundu, Sourav; Ari, Samit
• Is Part Of: Biomedical signal processing and control. Volume 55(2020)
• Publication Date: 2020-01
• Subjects: Brain–computer interface (BCI) -- Convolutional neural network (CNN) -- Ensemble of support vector machines -- Fisher ratio -- P300 speller; 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.2019.101645
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlights: In this paper, a novel method has been developed for the P300 speller system based on CNN and ESVM. Different combinations of CNN architectures are proposed to extract high-level features from EEG data automatically. F-ratio based technique has been proposed to select optimal features from the extracted features set for P300 based character recognition. Abstract: In this work, a brain–computer interface (BCI) system for character recognition has been proposed based on the P300 signal. P300 signal classification is the most challenging task in electroencephalography signal processing as it is affected by the surrounding noise and low signal-to-noise ratio (SNR). Feature extraction and feature selection are essential steps for any classification task. Most of the earlier techniques reported hand-crafted features for detection of P300 signal. However, the hand-crafted features are not efficient to represent the signal properly due to surrounding environment and subject variability. Motivated by this, convolutional neural network (CNN) has been proposed for automatic high-level feature extraction to detect P300 signal. In general, CNN model consists of convolutional and fully-connected layers followed by a softmax layer. In the developed system, two different convolutional layers are used to extract the spatial and temporal features from the dataset. Also, a 2D convolutional layer based CNN architecture has been proposed where spatio-temporal feature is extracted in a single layer. To mitigate the over-fitting problem, dropout is employed in CNN architecture, which improves the network performance. After extracting high-level features, Fisher ratio ( F-ratio ) based feature selection is proposed to find the optimal features. The optimal features are used in the ensemble of support vector machine (ESVM) classifier for P300 detection. ESVM has been adopted in this work to minimize the classifier variability. The models are tested on two widely used datasets, and the experimental results show better or comparable performance compared to the earlier reported techniques.