Research on feature enhancement algorithms for electromyographic signals under vibration disturbance for gesture intent recognition

Surface electromyography (sEMG) signals are widely employed in prosthetic control applications. However, their inherently weak amplitudes and high impedance render them highly susceptible to environmental disturbance, precluding stable control under adverse conditions such as vibration. To address this, this paper proposes an AE-ANN-based feature enhancement algorithm. At the feature level, it employs an autoencoder (AE) to reconstruct EMG signal features affected by vibration disturbance. An artificial neural network (ANN) is then introduced to supervise this reconstruction, generating an anti-disturbance model that provides feature inputs for gesture classification under vibration conditions, thereby improving gesture classification performance in the presence of vibrational disturbance. Experimental results demonstrate that within the frequency range corresponding to common electric hand tools, the proposed feature enhancement algorithm significantly improves the separability of the EMG feature space and gesture recognition accuracies, thereby providing a theoretical foundation for the practical application of this method.