Eeg emotion recognition based on channel spatio-temporal multi-dimensional feature extraction network

Electroencephalogram (EEG) signals capture the temporal and spatial characteristics of neuronal activities recorded by multiple electrodes and their data complexity is high. In order to effectively extract deeper emotional features from EEG signals, this paper proposes a Channel Spatio-temporal Multi-dimensional Feature Extraction Network (CSMFEN) for EEG emotion recognition. CSMFEN conducts feature learning on EEG signals from three dimensions which include channel domain, spatial domain and temporal domain. In the channel domain, CSMFEN models the relative importance of different EEG electrodes by assigning adaptive weights to each channel, thereby enhancing discriminative channel representations for emotion classification. In the spatial domain, an enhanced dynamic convolution mechanism is proposed to capture inter-channel spatial relationships, where the input signal space size information and convolution kernel weight information are jointly considered. Different attention mechanisms are employed to dynamically generate convolution kernels, enabling flexible spatial feature extraction. In the temporal domain, a Long Short-Term Memory (LSTM) network is utilized to learn the temporal dependencies and sequential dynamics of EEG signals. Experimental validation of CSMFEN is conducted on the SEED and DEAP datasets. The results demonstrate that CSMFEN can effectively learn discriminative representations and achieve competitive emotion recognition performance on both datasets.