Hand Gesture Recognition Using PSO-RF Based on sEMG Signals

Stroke and other conditions can cause damage to the neuromuscular information transmission pathways,thereby inducing limb motor dysfunction. Surface electromyography (sEMG) - based gesture recognition is of great significance in human - computer interaction for stroke rehabilitation. This paper proposes a multi - gesture recognition method based on particle swarm optimization (PSO) - optimized random forest (RF), referred to as PSO - RF. First, a 1 - Hz Butterworth filter is applied to the sEMG signals from the Ninapro DB1 dataset to preprocess the signals and remove noise. Then, time - domain, frequency - domain, and time - frequency - domain features are extracted using a 250 - ms sliding window. Finally, the parameters of RF are optimized by PSO to classify 12 types of gestures. Experimental results show that the average accuracy of PSO - RF is 94.15%, which is 4.4% higher than that of the unoptimized RF (89.75%). This method significantly improves the classification performance of multi - channel sEMG signals.