Identifying grasping motions from sEMG using non-negative matrix factorization and the Frisch scheme: a systematic, comparative analysis

Surface electromyography signals (sEMG) provide valuable insights into muscle activation patterns and are widely used in applications such as prosthetics, human-robot interaction, and rehabilitation. Extracting accurate muscle synergies from these signals is crucial for improving control and functionality in these systems. This study explores different methods for extracting muscle activation patterns from sEMG signals, focusing on Non-Negative Matrix Factorization (NMF), Sparse NMF (SNMF), the Frisch Scheme, and a hybrid approach combining Frisch scheme with NMF. The objective is to improve muscle synergy identification for applications in prosthetics and human-robot interaction. Results show that NMF-based methods effectively extract synergies, while the Frisch Scheme alone struggles to identify linear relationships without a reference. However, when combined with NMF, Frisch scheme improves noise reduction and signal reconstruction. Statistical analysis reveals significant differences between methods in single-grasping motion recognition, while multi-gesture datasets remain challenging due to (nonlinear) overlapping of activations. This research contributes to sEMG signal processing, highlighting the strengths and limitations of each method. Future work should focus on studying the use of the Frisch Scheme in greater detail, particularly in combination with other methods, to fully leverage its advantages.