Related Experiment Video
Updated: Mar 7, 2026
Capturing Dynamic Finger Gesturing with High-resolution Surface Electromyography and Computer Vision
Published on: March 28, 2025
Advancing individual finger classification through a sandwich enhanced CBAM network with ultra-high-density EEG data
Xinguo Zhang1, Yiman Zhang2, Hong Peng3
1Key Laboratory of China's Ethnic Languages and Information Technology of Ministry of Education, Chinese National Information Technology Research Institute, Northwest Minzu University, Lanzhou, China.
We developed a novel Sandwich enhanced Convolutional Block Attention Module (SCBAM) for ultra-high-density electroencephalography (uHD EEG) to decode individual finger movements. Our SCBAM model significantly improves classification accuracy for dexterous tasks in Brain-Computer Interfaces (BCI).
Area of Science:
- Neuroscience
- Biomedical Engineering
- Machine Learning
Background:
- Ultra-High-Density Electroencephalography (uHD EEG) shows promise for individual finger decoding.
- Accurate classification of subtle finger movements is challenging due to overlapping cortical activity.
- Standard neural network architectures struggle to isolate these fine-grained spatial features.
Purpose of the Study:
- To introduce a novel deep learning architecture, the Sandwich enhanced Convolutional Block Attention Module (SCBAM), for improved individual finger decoding using uHD EEG.
- To address the limitations of existing methods in capturing high-dimensional spatial features for precise movement classification.
- To explore the potential of SCBAM in advancing Brain-Computer Interface (BCI) applications for dexterous tasks.
Main Methods:
- Proposed the SCBAM, a hybrid network integrating dual attention mechanisms within a convolutional structure.
- Employed a unique 'sandwich' design to refine high-dimensional spatial features effectively.
- Conducted binary and five-class classification experiments across multiple subjects using uHD EEG data.
Main Results:
- Achieved an average accuracy of 78.63% in binary finger classification (highest 85% for Thumb vs. Ring).
- Attained an average accuracy of 61.12% in five-class finger classification (highest 62.36% for subject S2).
- Demonstrated superior performance compared to benchmark networks and traditional classifiers (SVM, MLP), validating the efficacy of SCBAM's attention mechanisms.
Conclusions:
- The SCBAM network offers a high-performance solution for individual finger classification using uHD EEG.
- This study addresses a gap in five-class finger classification research with HDEEG data.
- The findings highlight the potential of uHD EEG and advanced deep learning models like SCBAM for sophisticated BCI control.
