Abstract

Info

Title:: OEDCNet-based motor imagery electroencephalogram signal classification algorithm

Author(s):: ZHU Zhan; WANG Raofen; WU Jianzhen; School of Electronic and Electrical Engineering, Shanghai University of Engineering Science, Shanghai 201620, China

Keywords:: Keywords: electroencephalography signal motor imagery overlapping segmentation-recombination efficient channel attention dilated causal convolutional neural network

Abstract:: Abstract: To address the problems of insufficient sample size and cross-channel redundancy in motor imagery electroencephalogram (EEG), which leads to inadequate feature extraction in traditional models, a novel OEDCNet algorithm for motor imagery EEG signal classification is proposed. An overlapping segmentation-recombination approach is first introduced to optimize the data augmentation strategy. By subjecting the original EEG signals to overlapping segmentation and subsequent recombination, this approach expands the sample size while mitigating overfitting risks, thereby generating augmented sample data. Subsequently, an efficient channel attention mechanism is utilized to assign optimal weights to different channels and suppress interference from redundant channels. Finally, by improving the EEGNet approach, a dilated causal convolutional neural network is established to enable multi-scale feature extraction and long-range temporal dependency perception. The constructed OEDCNet model achieves an average accuracy of 77.08% and a Kappa value of 0.689 on the BCI Competition IV-2a dataset, outperforming other models. Experimental results demonstrate that the proposed OEDCNet method exhibits superior classification performance, thus providing a new technical pathway for the research on motor imagery EEG signal.