Brain visual image signal classification via hybrid dilation residual shrinkage network with spatio-temporal feature fusion

Brain–computer interface (BCI) technology based on electroencephalogram (EEG) has attracted widespread attention, among which interpretation, pattern recognition, and classification of brain activity through EEG are promising researches. However, EEG-based object classification is still confronted with enormous challenges in terms of the performance and interpretability of human brain signals. Accordingly, this paper constructs a novel hybrid dilation residual shrinkage network with Spatio-temporal feature fusion to research brain visual images classification. Inspired by visual attention and brain memory mechanisms, a hybrid dilation residual shrinkage module is designed to obtain the features of interest and reduce noise and redundant information. Then, EEG signals are encoded and stored in terms of the temporal and spatial dimensions, respectively. On the basis of the characteristics of the EEG signals, this work utilizes the gated recurrent unit network to generate temporal features and spatial features are obtained through a 2D hybrid dilation convolution module. Finally, the extracted spatio-temporal features are concatenated and then retrieved. Results indicate that the designed model is usable and effective. The proposed network achieves better classification performance compared with the existing methods.