Designing a Modified Grey Wolf Optimizer Based Cyclegan Model for Eeg Mi Classification in Bci

Published: Jan. 1, 2023

Language: Английский

---

TDLNet: Transfer Data Learning Network for Cross-Subject Classification Based on Multiclass Upper Limb Motor Imagery EEG

IEEE Transactions on Neural Systems and Rehabilitation Engineering, Journal Year: 2023, Volume and Issue: 31, P. 3958 - 3967

Published: Jan. 1, 2023

The limited number of brain-computer interface based on motor imagery (MI-BCI) instruction sets for different movements single limbs makes it difficult to meet practical application requirements. Therefore, designing a single-limb, multi-category (MI) paradigm and effectively decoding is one the important research directions in future development MI-BCI. Furthermore, major challenges MI-BCI difficulty classifying brain activity across individuals. In this article, transfer data learning network (TDLNet) proposed achieve cross-subject intention recognition multiclass upper limb imagery. TDLNet, Transfer Data Module (TDM) used process electroencephalogram (EEG) signals groups then fuse channel features through two one-dimensional convolutions. Residual Attention Mechanism (RAMM) assigns weights each EEG signal dynamically focuses channels most relevant specific task. Additionally, feature visualization algorithm occlusion frequency qualitatively analyze TDLNet. experimental results show that TDLNet achieves best classification datasets compared CNN-based reference methods method. 6-class scenario, obtained an accuracy 65%±0.05 UML6 dataset 63%±0.06 GRAZ dataset. demonstrate framework can produce distinct classifier patterns multiple categories frequencies. ULM6 available at https://dx.doi.org/10.21227/8qw6-f578.

Language: Английский

---

Dual selections based knowledge transfer learning for cross-subject motor imagery EEG classification

Frontiers in Neuroscience, Journal Year: 2023, Volume and Issue: 17

Published: Nov. 28, 2023

Introduction Motor imagery electroencephalograph (MI-EEG) has attracted great attention in constructing non-invasive brain-computer interfaces (BCIs) due to its low-cost and convenience. However, only a few MI-EEG classification methods have been recently applied BCIs, mainly because they suffered from sample variability across subjects. To address this issue, the cross-subject scenario based on domain adaptation widely investigated. existing often encounter problems such as redundant features incorrect pseudo-label predictions target domain. Methods achieve high performance classification, paper proposes novel method called Dual Selections Knowledge Transfer Learning (DS-KTL). DS-KTL selects both discriminative source corrects pseudo-labels The applies centroid alignment samples initially, then adopts Riemannian tangent space for feature adaptation. During adaptation, dual selections are performed with regularizations, which enhance during iterations. Results discussion Empirical studies conducted two benchmark datasets demonstrate feasibility effectiveness of proposed under multi-source single-target single-source strategies. achieves significant improvement similar efficiency compared state-of-the-art methods. Ablation also evaluate characteristics parameters method.

Language: Английский

---

Optimal Channel Selection of Multiclass Motor Imagery Classification Based on Fusion Convolutional Neural Network with Attention Blocks

Sensors, Journal Year: 2024, Volume and Issue: 24(10), P. 3168 - 3168

Published: May 16, 2024

The widely adopted paradigm in brain–computer interfaces (BCIs) involves motor imagery (MI), enabling improved communication between humans and machines. EEG signals derived from MI present several challenges due to their inherent characteristics, which lead a complex process of classifying finding the potential tasks specific participant. Another issue is that BCI systems can result noisy data redundant channels, turn increased equipment computational costs. To address these problems, optimal channel selection multiclass classification based on Fusion convolutional neural network with Attention blocks (FCNNA) proposed. In this study, we developed CNN model consisting layers multiple spatial temporal filters. These filters are designed specifically capture distribution relationships signal features across different electrode locations, as well analyze evolution over time. Following layers, Convolutional Block Module (CBAM) used to, further, enhance feature extraction. selection, genetic algorithm select set channels using new technique deliver fixed variable for all participants. proposed methodology validated showing 6.41% improvement compared most baseline models. Notably, achieved highest results 93.09% binary classes involving left-hand right-hand movements. addition, cross-subject strategy yielded an impressive accuracy 68.87%. was enhanced, reaching 84.53%. Overall, our experiments illustrated efficiency both classification, superior either full or reduced number channels.

Language: Английский

---

Dual-Branch Convolution Network With Efficient Channel Attention for EEG-Based Motor Imagery Classification

IEEE Access, Journal Year: 2024, Volume and Issue: 12, P. 74930 - 74943

Published: Jan. 1, 2024

Brain-Computer Interface (BCI) is a revolutionary technique that employs wearable electroencephalography (EEG) sensors and artificial intelligence (AI) to monitor decode brain activity. EEG-based motor imagery (MI) signal widely utilized in various BCI fields including intelligent healthcare, robot control, smart homes. Yet, the limited capability of decoding signals remains significant obstacle techniques expansion. In this study, we describe an architecture known as dual-branch attention temporal convolutional network (DB-ATCNet) for MI classification. DB-ATCNet improves classification performance with relatively fewer parameters by utilizing channel attention. The model consists two primary modules: convolution (ADBC) fusion (ATFC). ADBC module utilizes extract low-level MI-EEG features incorporates improve spatial feature extraction. ATFC sliding windows self-attention obtain high-level features, strategies minimize information loss. achieved subject-independent accuracies 87.33% 69.58% two-class four-class tasks, respectively, on PhysioNet dataset. On Competition IV-2a dataset, it accuracy 71.34% 87.54% subject-dependent evaluations, surpassing existing methods. code available at https://github.com/zk-xju/DB-ATCNet.

Language: Английский

---