A novel spatiotemporal filter for high-density EMG signals classification-based myoelectric control

Sophisticated prosthetic devices use classification algorithms to recognize electromyography (EMG) signal variations, accurately identifying user-intended motion gestures. Achieving high classification accuracy is crucial for the real-time control of prosthetic limbs. In this paper, we introduce an extended version of common spatial pattern (CSP) method by incorporating time delay embedding to extract spatial and temporal information into the classifier on myoelectric control performance. Nineteen able-bodied subjects in the experiment, and the experiment data from eight classes are used to classify based on the new spatiotemporal domain space proposed. The performance of the proposed approach surpasses that of traditional methods relying on time-domain features and autoregressive model and achieves a 10% increase in average classification accuracy (p < 0.05).