The Method of Spatiotemporal-Frequency Multidimensional Feature Fusion Based on Emotional EEG Signals

Emotion recognition technology is an important research direction in the field of artificial intelligence and emotional computing, and it has been widely used in the fields of mental health, education and medical care. As a direct reflection of emotional states, electroencephalogram (EEG) signals have become an important data source for emotion recognition research. Due to the low signal-to-noise ratio and non-stationary nature of EEG signals, they are easily affected by external interference, thus requiring a rigorous preprocessing procedure. Traditional research on multi-channel EEG signal feature extraction often focuses on the time-domain and frequency-domain characteristics of the signal itself, while neglecting the correlation features between brain regions. This leads to a more limited and single-dimensional feature extraction process. To address the above issues, we propose a dual segmentation of multi-channel EEG signals across both frequency bands and brain regions, combined with cross-integration of various time-frequency features. This approach enables the fusion of spatiotemporal-frequency multidimensional features of EEG signals. On this basis, we improved the traditional MLP neural network by introducing a regularization strategy. Experimental results validate the effectiveness of the proposed spatiotemporal-frequency multidimensional feature fusion method. Based on the EEG emotion dataset (SEED), cross-subject emotion classification experiments were conducted. Compared to baseline methods, the proposed approach achieved the best recognition performance, with an accuracy of 87.41 %. This demonstrates that the method can significantly improve the precision of EEG-based emotion decoding, exhibiting both high accuracy and good generalization capabilities.