Attention-fused residual transformer CNN for robust lower limb movement recognition

Detecting lower limb movements from surface electromyography (sEMG) signals has received more attention, because of its importance in prosthetic control, robotic applications and medical rehabilitation. sEMG signals offer a non-invasive and accurate method to recognize movement intent. Conventional...

Full description

Saved in:
Bibliographic Details
Main Authors: A. Anitha, D. Jeraldin Auxillia
Format: Article
Language:English
Published: Taylor & Francis Group 2025-07-01
Series:Automatika
Subjects:
Online Access:https://www.tandfonline.com/doi/10.1080/00051144.2025.2513734
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1849416414588305408
author A. Anitha
D. Jeraldin Auxillia
author_facet A. Anitha
D. Jeraldin Auxillia
author_sort A. Anitha
collection DOAJ
description Detecting lower limb movements from surface electromyography (sEMG) signals has received more attention, because of its importance in prosthetic control, robotic applications and medical rehabilitation. sEMG signals offer a non-invasive and accurate method to recognize movement intent. Conventional machine-learning approaches depend on manual feature extraction, which consumes more time and is susceptible to noise interference and class imbalance. To address these challenges, a new framework that combines an Attention-Fused Residual-Transformer Convolutional Neural Network (AF-RT-CNN) is proposed. Data augmentation is applied to create more samples for minority classes, addressing class imbalance problems and improving recognition reliability. The AF-RT-CNN architecture combines residual blocks, attention mechanism and Transformer Encoder aiding robust feature extraction, good generalization capability and pattern recognition. The proposed framework was evaluated across 11 healthy individuals and 11 patients with lower limb impairments, across three distinct categories of lower limb movements sit, stand and gait. The method achieved a remarkable accuracy of 99.35% for the healthy group and 99.54% for the pathological group. When combining both groups, the overall accuracy reached 98.74%. The results indicate the effectiveness of the proposed approach in rehabilitation and assistive technology for lower limb motor control.
format Article
id doaj-art-ce98a8efc6b74cb08aed2f15ef168af0
institution Kabale University
issn 0005-1144
1848-3380
language English
publishDate 2025-07-01
publisher Taylor & Francis Group
record_format Article
series Automatika
spelling doaj-art-ce98a8efc6b74cb08aed2f15ef168af02025-08-20T03:33:11ZengTaylor & Francis GroupAutomatika0005-11441848-33802025-07-0166352854210.1080/00051144.2025.2513734Attention-fused residual transformer CNN for robust lower limb movement recognitionA. Anitha0D. Jeraldin Auxillia1Department of Electronics and Communication Engineering, St. Xavier’s Catholic College of Engineering, Nagercoil, Tamil Nadu, IndiaDepartment of Electronics and Communication Engineering, St. Xavier’s Catholic College of Engineering, Nagercoil, Tamil Nadu, IndiaDetecting lower limb movements from surface electromyography (sEMG) signals has received more attention, because of its importance in prosthetic control, robotic applications and medical rehabilitation. sEMG signals offer a non-invasive and accurate method to recognize movement intent. Conventional machine-learning approaches depend on manual feature extraction, which consumes more time and is susceptible to noise interference and class imbalance. To address these challenges, a new framework that combines an Attention-Fused Residual-Transformer Convolutional Neural Network (AF-RT-CNN) is proposed. Data augmentation is applied to create more samples for minority classes, addressing class imbalance problems and improving recognition reliability. The AF-RT-CNN architecture combines residual blocks, attention mechanism and Transformer Encoder aiding robust feature extraction, good generalization capability and pattern recognition. The proposed framework was evaluated across 11 healthy individuals and 11 patients with lower limb impairments, across three distinct categories of lower limb movements sit, stand and gait. The method achieved a remarkable accuracy of 99.35% for the healthy group and 99.54% for the pathological group. When combining both groups, the overall accuracy reached 98.74%. The results indicate the effectiveness of the proposed approach in rehabilitation and assistive technology for lower limb motor control.https://www.tandfonline.com/doi/10.1080/00051144.2025.2513734Lower limbsEMGdeep learningResNettransformer
spellingShingle A. Anitha
D. Jeraldin Auxillia
Attention-fused residual transformer CNN for robust lower limb movement recognition
Automatika
Lower limb
sEMG
deep learning
ResNet
transformer
title Attention-fused residual transformer CNN for robust lower limb movement recognition
title_full Attention-fused residual transformer CNN for robust lower limb movement recognition
title_fullStr Attention-fused residual transformer CNN for robust lower limb movement recognition
title_full_unstemmed Attention-fused residual transformer CNN for robust lower limb movement recognition
title_short Attention-fused residual transformer CNN for robust lower limb movement recognition
title_sort attention fused residual transformer cnn for robust lower limb movement recognition
topic Lower limb
sEMG
deep learning
ResNet
transformer
url https://www.tandfonline.com/doi/10.1080/00051144.2025.2513734
work_keys_str_mv AT aanitha attentionfusedresidualtransformercnnforrobustlowerlimbmovementrecognition
AT djeraldinauxillia attentionfusedresidualtransformercnnforrobustlowerlimbmovementrecognition