An end-to-end residual convolutional Kolmogorov-Arnold Network framework for robust gait phase recognition from surface EMG
Accurate recognition of gait phases from surface electromyography (sEMG) signals is essential for intelligent prosthetics, exoskeletons, and rehabilitation. This study proposes a residual convolutional Kolmogorov-Arnold network (RCKAN) that integrates convolutional feature encoding, ridge-regularized nonlinear mapping, and residual learning for five-phase gait classification. Experimental results on data from 40 healthy subjects demonstrate that RCKAN achieves state-of-the-art accuracy of 95.52%, surpassing conventional machine learning and advanced deep learning architectures. The model maintains strong performance in transitional phases such as heel strike and toe-off, confirming its robustness and potential for real-time gait analysis and assistive control applications.
Authors
- Shuqi Zhu
- Yongxiang Zou (ORCID: https://orcid.org/0000-0003-1521-4864)
- Xiuxiu Shi
- Jinghui Guo (ORCID: https://orcid.org/0009-0002-9159-2840)
Institutions
- Chinese Academy of Sciences (CN)
- Nanjing University of Information Science and Technology (CN)
- Chinese PLA General Hospital (CN)
- Zhengzhou University (CN)
- Institute of Automation (CN)
- Henan Cancer Hospital (CN)
- University of Chinese Academy of Sciences (CN)
Publication Details
- Journal
- Computer Methods in Biomechanics & Biomedical Engineering
- Published
- 2026-11-18
- DOI
- https://doi.org/10.1080/10255842.2026.2755547
- Primary Topic
- Muscle activation and electromyography studies
- Type
- article
- Field-Weighted Citation Impact
- 0.00
Funders
- Natural Science Foundation for Young Scientists of Shanxi Province