Jove
Visualize
Contact Us
JoVE
x logofacebook logolinkedin logoyoutube logo
ABOUT JoVE
OverviewLeadershipBlogJoVE Help Center
AUTHORS
Publishing ProcessEditorial BoardScope & PoliciesPeer ReviewFAQSubmit
LIBRARIANS
TestimonialsSubscriptionsAccessResourcesLibrary Advisory BoardFAQ
RESEARCH
JoVE JournalMethods CollectionsJoVE Encyclopedia of ExperimentsArchive
EDUCATION
JoVE CoreJoVE BusinessJoVE Science EducationJoVE Lab ManualFaculty Resource CenterFaculty Site
Terms & Conditions of Use
Privacy Policy
Policies

Related Experiment Videos

sEMG gait phase classification based on CNN-transformer and transfer learning.

Qixin Guo1, Yi Zheng1,2, Chang Li1

  • 1School of Mechanical and Automotive Engineering, Qingdao University of Technology, Qingdao, China.

Computer Methods in Biomechanics and Biomedical Engineering
|July 9, 2026
PubMed
Summary
This summary is machine-generated.

Related Concept Videos

You might also read

Related Articles

Articles linked to this work by shared authors, journal, and citation graph.

Sort by
Same author

PulseGAN: Learning to Generate Realistic Pulse Waveforms in Remote Photoplethysmography.

IEEE journal of biomedical and health informatics·2021
Same author

Dexmedetomidine post-conditioning ameliorates long-term neurological outcomes after neonatal hypoxic ischemia: The role of autophagy.

Life sciences·2021
Same author

[Corrigendum] Ski prevents TGF‑β‑induced EMT and cell invasion by repressing SMAD‑dependent signaling in non‑small cell lung cancer.

Oncology reports·2021
Same author

Synergistic Effects of Alpha Olefin Sulfonate and Sodium Alginate on Inkjet Printing of Cotton/Polyamide Fabrics.

Langmuir : the ACS journal of surfaces and colloids·2021
Same author

Diagnostic value of microRNA-25 in patients with non-small cell lung cancer in Chinese population: A systematic review and meta-analysis.

Medicine·2020
Same author

Development and external validation of a COVID-19 mortality risk prediction algorithm: a multicentre retrospective cohort study.

BMJ open·2020

This study introduces a lightweight CNN-Transformer model for recognizing motion intentions in lower-limb exoskeleton rehabilitation. The advanced model achieves high accuracy and fast inference, offering a robust, patient-adaptive solution for real-time control.

Area of Science:

  • Biomedical Engineering
  • Rehabilitation Robotics
  • Machine Learning for Healthcare

Background:

  • Accurate motion intention recognition is crucial for effective control of lower-limb rehabilitation exoskeletons.
  • Existing methods often struggle with inter-subject variability and real-time processing demands.
  • Surface electromyography (sEMG) signals are key for detecting user intent.

Purpose of the Study:

  • To develop a lightweight and accurate model for real-time motion intention recognition in exoskeleton control.
  • To address challenges posed by inter-subject variability in sEMG data.
  • To enable personalized adaptation for improved rehabilitation outcomes.

Main Methods:

  • A hybrid Convolutional Neural Network (CNN)-Transformer model was designed with a small parameter count (0.35M).
Keywords:
Lower-limb exoskeletongait phase recognitionintention recognitionsEMG

Related Experiment Videos

  • Inverted residual CNNs were used for extracting multi-dimensional local features from sEMG signals.
  • Transformer modules were employed to capture long-range dependencies within the data.
  • A two-stage transfer learning approach, including source-domain pre-training and target-domain fine-tuning, was implemented for personalization.

    • Main Results:

    • The proposed model achieved a low inference latency of 5.6 ms.
    • High accuracies were reported: 93.46% for single-subject recognition and 92.97% for cross-subject recognition.
    • The model demonstrated effectiveness in handling inter-subject variability through personalized adaptation.

    Conclusions:

    • The lightweight CNN-Transformer hybrid model offers a robust and patient-adaptive solution for real-time exoskeleton control.
    • The two-stage transfer learning strategy effectively addresses inter-subject variability.
    • This approach significantly advances the potential for personalized lower-limb exoskeleton rehabilitation.