15 May 2023 in IEEE Transactions on Biomedical Engineering
Objective : Electroencephalogram (EEG) signal recognition based on deep learning technology requires the support of sufficient data. However, training data scarcity usually occurs in subject-specific motor imagery tasks unless multisubject data can be used to enlarge training data. Unfortunately, because of the large discrepancies between data distributions from different subjects, model performance could only be improved marginally or even worsened by simply training on multisubject data. Method : This paper proposes a novel weighted multi-branch (WMB) structure for handling multisubject data to solve the problem, in which each branch is responsible for fitting a pair of source-target subject data and adaptive weights are used to integrate all branches or select branches with the largest weights to make the final decision. The proposed WMB structure was applied to six well-known deep learning models (EEGNet, Shallow ConvNet, Deep ConvNet, ResNet, MSFBCNN, and EEG_TCNet) and comprehensive experiments were conducted on EEG datasets BCICIV-2a, BCICIV-2b, high gamma dataset (HGD) and two supplementary datasets. Result : Superior results against the state-of-the-art models have demonstrated the efficacy of the proposed method in subject-specific motor imagery EEG classification. For example, the proposed WMB_EEGNet achieved classification accuracies of 84.14%, 90.23%, and 97.81% on BCICIV-2a, BCICIV-2b and HGD, respectively. Conclusion : It is clear that the proposed WMB structure is capable to make good use of multisubject data with large distribution discrepancies for subject-specific EEG classification.
18 September 2024 in IEEE Journal of Biomedical and Health Informatics
For privacy protection of subjects in electroencephalogram (EEG)-based brain-computer interfaces (BCIs), using source-free domain adaptation (SFDA) for cross-subject recognition has proven to be highly effective. However, updating and storing a model trained on source subjects for each new subject can be inconvenient. This paper extends Euclidean alignment (EA) to propose adaptive Euclidean alignment (AEA), which learns a projection matrix to align the distribution of the target subject with the source subjects, thus eliminating domain drift issues and improving model classification performance of subject-independent BCIs. Combining the proposed AEA with various existing SFDA methods, such as SHOT, GSFDA, and NRC, this paper presents three new methods: AEA-SHOT, AEA-GSFDA, and AEA-NRC. In our experimental studies, these AEA-based SFDA methods were applied to four well-known deep learning models (i.e., EEGNet, Shallow ConvNet, Deep ConvNet, and MSFBCNN) on two motor imagery (MI) datasets, one event-related potential (ERP) dataset and one steady-state visual evoked potentials (SSVEP) dataset. The advanced cross-subject EEG classification performance demonstrates the efficacy of our proposed methods. For example, AEA-SHOT achieved the best average accuracy of 81.4% on the PhysioNet dataset.
18 July 2019 in IEEE Access
Currently, majority of person re-identification (reID) technologies are network-constrained by Dropout regularization, which relies on the random zeroing out of some features to make these features more independent. However, such random zeroing regularization methods are not very effective for improving network performance, because they neglect the unique contribution of different features to the network performance. To improve the value of indiscriminative features in network training, a DropEasy-based person reID method is proposed in this paper. Features are classified into discriminative and indiscriminative ones, according to the distance between the feature vectors of positive or negative sample pairs wherein the discriminative features are zeroed out, while the indiscriminative features are reserved, and the network only learns through indiscriminative features. Furthermore, because networks are always inclined to make up for incomplete information by drawing on the surrounding features in the feature maps, Dropout loses its effectiveness for the network-constraints. To solve this challenge, the DropEasy2d method that can be effectively applied to convolutional layers is further proposed in this paper. DropEasy2d searches discriminative feature areas in the feature maps by sliding and zeroing windows while reserving the indiscriminative features areas to constrain network learning. The effectiveness of the proposed method is demonstrated using the Market-1501, DuckMTMC-reID, and CUHK03 datasets. For example, in the Market-1501 dataset, DropEasy can improve the mean average precision (mAP) and Rank-1 accuracy of the ID-discriminative embedding (IDE) to 72.7%(+8.8%) and 90.5%(+6.8%), respectively, while DropEasy2d can raise them to 68.5%(+4.6%) and 88.7%(+5.0%), respectively. Based on the results, the proposed method can improve the network performance during the extraction and generalization of the discriminative features.
15 June 2026 in Biomedical Signal Processing and Control