Electromyogram decomposition using the single-lineage clustering algorithm and wavelets

The decomposition of intramuscular myoelectric signals (EMG signals) can be considered as a classification problem, where both supervised and unsupervised classification techniques have to be used. Due to the low SNR and due to the lack of a priori knowledge, the unsupervised classification needs a lot of interactive tasks. In this paper, we describe the case when wavelet coefficients from selected frequency bands improve the performance of the unsupervised classification algorithm. Furthermore, we compare a wavelet-based distance measure with a commonly-used distance measure in the context of the single-linkage clustering algorithm. Tests with EMG recordings yield very good results for the wavelet-based distance measure