2 June 2022 in IEEE Transactions on Affective Computing
Automatic emotion recognition based on electroencephalogram (EEG) has attracted rapidly increasing interests. Due to large inter-subject variabilities, subject-independent emotion recognition faces great challenges. Recently, domain adaptation methods have been successfully applied in this field due to their ability to align features from different subjects. However, since EEG signals corresponding to some emotions have similar oscillation patterns, they are often confused and aligned to the wrong categories, which limits the generalization ability of the model across subjects. Besides, almost all methods only support offline applications, which require collecting a large number of samples of new subjects. To achieve online recognition, a simpler model is needed. In this paper, a novel Gated Recurrent Unit-Minimum Class Confusion (GRU-MCC) model is proposed. Specifically, a simple feature extractor based on gated recurrent unit (GRU) is firstly applied to model the spatial dependence of multiple electrodes and obtain high-level discriminative features. Then, during training, minimum class confusion (MCC) loss is introduced to reduce the confusion between the correct and ambiguous classes for the target subject and increase the transfer gains. We conduct both offline and online experiments on two public datasets: SEED and MPED. The results indicate that our method can obtain the superior performance.
29 December 2023 in IEEE Transactions on Instrumentation and Measurement
Electroencephalogram (EEG) obtained from the scalp is often polluted with various artifacts. Deep learning (DL) has been preliminarily explored in EEG denoising and proved to be more promising than traditional methods. However, existing DL methods focus on recovering clean EEG, while ignoring the recovery of artifacts present in contaminated signals. This limits their ability to eliminate artifacts. In this paper, we propose a dual-branch interactive fusion neural network, called EEGIFNet, to remove artifacts from single-channel EEG. In EEGIFNet, the two branches are employed to recover clean EEG and residual artifacts respectively. Convolution layers and bidirectional gated recurrent unit (BiGRU) are cascaded in each branch to capture local and global features. We also introduce a interaction module to extract residual components from opposite branches, which benefits both branches from each other. In addition, a fusion module is designed to adaptively fuse the outputs of the two branches and finally reconstruct the clean EEG. We evaluate the proposed method on publicly available datasets including electrooculogram (EOG), electromyography (EMG), electrocardiogram (ECG), power line interference (PLI) and motion artifacts. Extensive experiments indicate that EEGIFNet outperforms the state-of-the-art methods on diverse artifacts and signal-to-noise (SNR) levels.
17 October 2024 in Computers in Biology and Medicine
4 April 2026 in Advanced Engineering Informatics
20 March 2025 in IEEE Transactions on Biomedical Engineering
Generalized zero-shot learning (GZSL) networks offer promising avenues for the development of user-friendly steady-state visual evoked potential (SSVEP) based brain-computer interfaces (BCIs), aiming to alleviate the training burden on users. These networks only require the user to provide training data from partial classes during training, yet they demonstrate the capability to classify all classes during testing. However, these GZSL networks have a large number of trainable parameters, resulting in long training times and difficulty to practicalize. In this study, we proposed a dual-attention structure to construct a lightweight GZSL network, termed GZSL-Lite. We first embedded the input training data-constructed class templates, manually constructed sine templates, and electroencephalogram (EEG) signals using convolution-based networks. The embedding part uses the same network weights to embed the features across different stimulus frequencies while reducing the depth of the network. After embedding, two branches of the dual-attention use class and sine templates to guide the feature extraction of the EEG signal with the attention mechanism, respectively. Compared to the networks that extract all features equally, dual-attention focuses only on EEG features relative to templates, which helps classification with fewer parameters. Finally, we used depthwise convolutional blocks to output classification results. Experimental evaluations conducted on two publicly available datasets demonstrate the efficacy of the proposed network. Comparative analysis reveals a remarkable reduction in trainable parameters to less than 1% of the SOTA counterpart, concurrently showing significant performance improvement.
1 November 2026 in IEEE Transactions on Biomedical Engineering
Data augmentation is important for enhancing subject-independent classification in deep learning (DL) approaches for steady-state visual evoked potential (SSVEP) brain-computer interfaces (BCIs) using electroencephalography (EEG). However, current augmentation techniques often inadequately exploit individual-specific style characteristics, limiting the model's robustness against inter-subject style variability. To tackle this problem, this study proposes a novel data augmentation method called Simultaneous Spatial-Energy Representation (SSER). SSER employs singular value decomposition (SVD) to extract spatial and energy representations from EEG signals, effectively capturing style characteristics. These representations are independently mixed across source domains during signal reconstruction, generating novel domains that cover a broader range of styles. This strategy promotes the learning of domain-invariant features and enhances the model's robustness to style variability. Comprehensive experiments on public datasets demonstrate that SSER outperforms state-of-the-art data augmentation techniques and generalizes well across various DL models. Furthermore, self-collected offline and online experiments involving 30 subjects provide additional evidence of the method's effectiveness. By simultaneously manipulating spatial and energy representations, SSER offers a richer characterization of EEG signal style variability, leading to superior performance. The proposed innovative data augmentation method advances subject-independent classification, facilitating the broader application of EEG-based BCIs in real-world scenarios.