One-dimensional CNN-based classification of fine hand movements using low-spatial-resolution sEMG

The control of myoelectric prostheses to perform fine finger movements represents a significant challenge in rehabilitation engineering , especially when using a reduced number of surface electromyography (sEMG) channels to ensure the clinical viability of the device. This work proposes and evaluates the performance of a one-dimensional Convolutional Neural Network (CNN) architecture for the classification of seven fine hand gestures from sEMG data acquired with a low-spatial-resolution system of only three channels. Data were collected from five healthy volunteers , pre-processed through filtering and overlapping windowing , and used to train the model. The CNN achieved a mean validation accuracy of 96.46% ± 1.31% , demonstrating high efficacy and robustness against inter-subject variability. The result indicates that the deep learning approach can extract discriminative features from complex sEMG signals , even with limited spatial information, presenting itself as a high-performance and clinically viable solution for the development of more accessible and intuitive prosthetic control systems.