A Lightweight Network with Domain Adaptation for Motor Imagery Recognition DOI Creative Commons
Xinmin Ding, Zenghui Zhang, Kun Wang

et al.

Entropy, Journal Year: 2024, Volume and Issue: 27(1), P. 14 - 14

Published: Dec. 27, 2024

Brain-computer interfaces (BCI) are an effective tool for recognizing motor imagery and have been widely applied in the control assistive operation domains. However, traditional intention-recognition methods face several challenges, such as prolonged training times limited cross-subject adaptability, which restrict their practical application. This paper proposes innovative method that combines a lightweight convolutional neural network (CNN) with domain adaptation. A feature extraction module is designed to extract key features from both source target domains, effectively reducing model's parameters improving real-time performance computational efficiency. To address differences sample distributions, adaptation strategy introduced optimize alignment. Furthermore, adversarial employed promote learning of domain-invariant features, significantly enhancing generalization ability. The proposed was evaluated on fNIRS dataset, achieving average accuracy 87.76% three-class classification task. Additionally, experiments were conducted two perspectives: model structure optimization data selection. results demonstrated potential advantages this applications recognition systems.

Language: Английский

Accurate identification of anxiety and depression based on the dlPFC in an emotional autobiographical memory task: A machine learning approach DOI
Guixiang Wang, Yusen Huang, Yan Zhang

et al.

Biomedical Signal Processing and Control, Journal Year: 2025, Volume and Issue: 104, P. 107503 - 107503

Published: Jan. 18, 2025

Language: Английский

Citations

0
Motor Imagery Classification Using fNIRS Brain Signals: A Method Based on Synthetic Data Augmentation and Cosine‐Modulated Attention DOI Creative Commons

Cheng Peng,

Baojiang Li, Haiyan Wang

et al.

Computational Intelligence, Journal Year: 2025, Volume and Issue: 41(2)

Published: March 21, 2025

ABSTRACT Functional near‐infrared spectroscopy (fNIRS), renowned for its high spatial resolution, shows substantial promise in brain‐computer interface (BCI) applications. However, challenges such as lengthy data acquisition processes and susceptibility to noise can limit availability reduce classification accuracy. To overcome these limitations, we introduce the CosineGAN‐transformer network (CGTNet), which integrates a dual discriminator GAN generating high‐quality synthetic with Transformer‐based network. Equipped multi‐head self‐attention mechanism, this excels at capturing intricate spatiotemporal relationships inherent high‐resolution fNIRS signals. The framework ensures that both temporal aspects of closely resemble original signals, thereby enhancing diversity fidelity. Experimental results on publicly available dataset, comprising 30 participants performing motor imagery tasks (right‐hand tapping, left‐hand foot tapping), demonstrate CGTNet achieves an accuracy 82.67%, outperforming existing methods. Key contributions work include use refined feature extraction Generative Adversarial Networks (GAN) maintains quality consistency. These advancements significantly improve robustness BCI systems, offering promising applications neurorehabilitation assistive technologies.

Language: Английский

Citations

0
Enhancing Classification Accuracy with Integrated Contextual Gate Network: Deep Learning Approach for Functional Near-Infrared Spectroscopy Brain–Computer Interface Application DOI Creative Commons
Jamila Akhter, Noman Naseer, Hammad Nazeer

et al.

Sensors, Journal Year: 2024, Volume and Issue: 24(10), P. 3040 - 3040

Published: May 10, 2024

Brain–computer interface (BCI) systems include signal acquisition, preprocessing, feature extraction, classification, and an application phase. In fNIRS-BCI systems, deep learning (DL) algorithms play a crucial role in enhancing accuracy. Unlike traditional machine (ML) classifiers, DL eliminate the need for manual extraction. neural networks automatically extract hidden patterns/features within dataset to classify data. this study, hand-gripping (closing opening) two-class motor activity from twenty healthy participants is acquired, integrated contextual gate network (ICGN) algorithm (proposed) applied that enhance classification The proposed extracts features filtered data generates patterns based on information previous cells network. Accordingly, performed similar generated dataset. accuracy of compared with long short-term memory (LSTM) bidirectional (Bi-LSTM). ICGN yielded 91.23 ± 1.60%, which significantly (p < 0.025) higher than 84.89 3.91 88.82 1.96 achieved by LSTM Bi-LSTM, respectively. An open access, three-class (right- left-hand finger tapping dominant foot tapping) 30 subjects used validate algorithm. results show can be efficiently two- problems fNIRS-based BCI applications.

Language: Английский

Citations

2
TPat: Transition pattern feature extraction based Parkinson’s disorder detection using FNIRS signals DOI
Türker Tuncer, İrem Taşçı, Burak Taşçı

et al.

Applied Acoustics, Journal Year: 2024, Volume and Issue: 228, P. 110307 - 110307

Published: Sept. 27, 2024

Language: Английский

Citations

2
A Lightweight Network with Domain Adaptation for Motor Imagery Recognition DOI Creative Commons
Xinmin Ding, Zenghui Zhang, Kun Wang

et al.

Entropy, Journal Year: 2024, Volume and Issue: 27(1), P. 14 - 14

Published: Dec. 27, 2024

Brain-computer interfaces (BCI) are an effective tool for recognizing motor imagery and have been widely applied in the control assistive operation domains. However, traditional intention-recognition methods face several challenges, such as prolonged training times limited cross-subject adaptability, which restrict their practical application. This paper proposes innovative method that combines a lightweight convolutional neural network (CNN) with domain adaptation. A feature extraction module is designed to extract key features from both source target domains, effectively reducing model's parameters improving real-time performance computational efficiency. To address differences sample distributions, adaptation strategy introduced optimize alignment. Furthermore, adversarial employed promote learning of domain-invariant features, significantly enhancing generalization ability. The proposed was evaluated on fNIRS dataset, achieving average accuracy 87.76% three-class classification task. Additionally, experiments were conducted two perspectives: model structure optimization data selection. results demonstrated potential advantages this applications recognition systems.

Language: Английский

Citations

0