.png){kind=icon}
.png){kind=icon}
Return
A dual-stage deep-CNN driven approach for resolving sensor displacement impact in EMG-PR systems
.png){kind=icon}
OA
AI
F
M
O
D
P
X
Y
G
DOI：10.1016/j.bspc.2026.111223.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Electrode displacement is a critical issue that degrades the performance of myoelectric control systems. Despite years of research towards resolving this issue, existing solutions are considered inadequate partly because they are not robust to the dynamics associated with electrode displacement. Thus, this study proposed a novel approach driven by Dual-Stage convolutional neural network (DS-CNN) for shift mitigation in myo-control systems. The DS-CNN comprises of two sequential stages: Stage 1, detects the occurrence of an electrode shift, and based on that, stage 2 initiates a specialized CNN to accurately decode the intended motion of the user. This accurate decoding is crucial for achieving robust real-time control of multifunctional prostheses and other rehabilitation systems. This method takes as input raw myoelectric signals, reducing the need for extensive preprocessing that are typically required in traditional machine learning-based pattern recognition approaches. The efficacy of the DS_CNN was evaluated with public dataset from 18 able-bodied individuals and in-house dataset from 4 amputee subjects, encompassing four distinct shift conditions (with shifts ranging from 7.50 mm to 20 mm). Extensive experimentation showed that the DS-CNN can adequately handle both transverse and longitudinal shifts, with an average increment in limb motion decoding accuracy of approximately 12.93% for the public dataset and 25.66% for the in-house dataset in comparison to state-of-the-art methods. Additionally, the DS-CNN recorded a performance that is nearly on par with the No-Shift benchmark scenario. Findings from this study offer novel insights for development of robust solutions for electrode displacement mitigation, thereby enhancing the overall performance of myoelectric systems.
Keywords:
Deep Learning Model
Pattern Recognition
Electromyogram
Prosthesis
Electrode Displacement
Journal
IF:
4.9
Papers:
9.7K
Citations:
2.4W
Organization
Cited Papers