The NR-BA-BFE-Net model incorporating MIMO noise reduction and band-wise attention improved emotion recognition accuracy by 3.27% over BFE-Net (95.56% vs 92.29%) in subject-independent EEG emotion classification on the SEED dataset (p=0.032).
A novel noise-robust band-attention BFE-Net framework improves EEG-based emotion recognition accuracy compared to state-of-the-art methods.
Effect estimate: 3.27% absolute improvement in accuracy vs BFE-Net
Absolute Event Rate: 95.56% vs 92.29%
p-value: p=0.032
Electroencephalogram (EEG) signals serve as a primary input for brain–computer interface (BCI) systems, and extensive research has been conducted on EEG-based emotion recognition. However, because EEG signals are inherently contaminated with various types of noise, the performance of emotion recognition is often degraded. Furthermore, the use of a Band Feature Extraction Neural Network (BFE-Net), a state-of-the-art (SOTA) method in this field, has limitations with respect to independent band-wise feature extraction and a simplistic band aggregation process to obtain final classification results. To address these problems, this study proposes the noise-robust band-attention BFE-Net framework, aiming to improve the conventional BFE-Net from two perspectives. First, we implement multiple-input, multiple-output (MIMO)-based preprocessing. Specifically, we utilize multichannel minima-controlled recursive averaging for precise non-stationary noise covariance estimation and generalized eigenvalue decomposition for subspace filtering to enhance the signal-to-noise ratio. Second, we propose an attention-based band aggregation mechanism. By integrating a band-wise self-attention mechanism, the model learns dynamic inter-band dependencies for more sophisticated feature fusion for classification. Experimental results on the SEED and SEED-IV datasets under a subject-independent protocol show that our model outperforms the SOTA BFE-Net by 3.27% and 3.34%, respectively. This confirms that rigorous MIMO noise reduction, combined with frequency-centric attention, significantly enhances the reliability and generalization of BCI systems.
Ji et al. (2026) studied Young adults (average age 23.27 years, 7 males and 8 females) undergoing EEG-based emotion recognition with multichannel EEG capturing emotional states induced by video clips (n=15). NR-BA-BFE-Net (Noise-Robust Band-Attention BFE-Net) incorporating MIMO noise reduction and band-wise self-attention feature fusion vs. State-of-the-art models including BFE-Net, SVM, DAMGCN, DGCNN, RGNN, TANN, BiHDM, GMSS, SOGNN and variants without noise reduction or attention was evaluated on Classification accuracy of EEG-based emotion recognition under subject-independent protocol (3.27% absolute improvement in accuracy vs BFE-Net, p=0.032). The NR-BA-BFE-Net model incorporating MIMO noise reduction and band-wise attention improved emotion recognition accuracy by 3.27% over BFE-Net (95.56% vs 92.29%) in subject-independent EEG emotion classification on the SEED dataset (p=0.032).