Tensor ring subspace analysis method for hand movement classification from multichannel surface EMG signals

Abstract

Tensor ring (TR) decomposition is a linear combination of tensor train (TT) decomposition. As a circular-dimensional permutation-invariant tensor-decomposition model, it yields more powerful and general low-rank representations of multiway data with great potential for a variety of applications in machine learning and signal processing. Motivated by these applications, in this study, we extend the TT-based EMG signal classification strategy, which was introduced in our conference paper from ICCS 2023, to a more general and efficient version that takes advantage of the TR model. By combining it with tensor subspace analysis (TSA), which additionally allows us to extract more discriminant 2D features, we demonstrate that the proposed method outperforms many competitive approaches for the classification of multichannel sEMG signals registered during various hand movements.