SEVEN-CLASS LOWER-LIMB ACTIVITY RECOGNITION USING THREE-CHANNEL SEMG

Three-channel sEMG recordings may contain different signatures of lower-limb activity, but separating movements is less straightforward when they recruit some of the same muscles. We examined this problem in seven right-leg activities using the conditioned envelopes recorded over rectus femoris, vastus lateralis, and semitendinosus. The experiment comprised five repeated movements and two sustained postures, each recorded for 40 s during one session with a 25-year-old participant without amputation. We divided the recordings into 300-ms windows with 60% overlap, extracted seven time-domain descriptors, and used them in a support vector machine (SVM). The stored test matrix contains 464 windows and 401 correct assignments, corresponding to 86.42% accuracy. The findings describe the discrimination achieved with the selected channels and indicate aspects of electrode placement and cycle labeling that warrant further testing.