Estimating Finger Joint Angles from Surface Electromyography Using Convolutional Neural Networks

This study proposes a novel method to improve the accuracy of finger joint angle estimation based on convolutional neural networks (CNNs) and surface electromyography (sEMG). sEMG, a complex biological signal measuring muscle electrical activity, plays a crucial role in controlling prosthetic devices for amputees. To address the challenge of accurately simulating continuous multi-finger movements in intelligent prosthetics, we introduce a CNN-based approach that improves joint angle prediction. Key innovations include the simultaneous estimation of five joint angles using a CNN regression model and the extraction of two time-domain and one frequency-domain features from sEMG signals. The method was validated using forearm sEMG data from eight channels and finger joint movement angles from six subjects performing nine distinct finger movements. The proposed method achieved high accuracy, with R2 values of 0.802, 0.805, 0.904, 0.931, and 0.899 for estimating the five finger joint angles, demonstrating its potential for enhancing multi-finger movement control in prosthetic hands.