Inception-CNN For Improvement Of P300 Detection In Brain-Computer Interface

P300 detection is a difficult task in the P300 speller-based brain-computer interface(BCI)system due to its low signal-to-noise ratio(SNR).Previously Convolutional Neural Networks(CNN)have shown prominent results for P300 detection as compared to various machine learning models.However,current CNN architectures limit P300 detection accuracy because these models usually only extract single-scale features.The Inception module has been proposed by Google Net to improve the classification ability of CNN using a multi-kernel convolutional layer.Aiming to enhance the P300 detection accuracy,the proposed model in this paper employed an inception module to traditional CNN architecture,which increases the width of a network.The proposed P300 detection model can extract multi-scale features using filters of different sizes to efficiently improve the P300 detection accuracy.Furthermore,the proposed model effectively learns discriminative spatial and temporal features of P300.The proposed P300 detection model was evaluated on dataset II of the third BCI competition.The experimental results show the effectiveness of the proposed model.More importantly,the character spelling accuracy over early epochs is greatly improved as compared to existing CNN architectures.