Hand Gesture Recognition of Double-Channel EMG Signals Based on Sample Entropy and PSO-SVM

Abstract

Considering wearable device requirements, such as injury-free, comfortable and light, a new method based on sample entropy and PSO-SVM for double-channel EMG signals recognition is proposed. First, the signal is segmented by using a non-overlapping adjacent sliding window. And the sample entropy of the valid information after signal in each window is decomposed by EMD is fast calculated to obtain a feature vector. Then the paper improved the PSO algorithm from two aspects. On the one hand, using a dynamic topology improved the impact of “premature” by replacing the global optimal value with neighborhood optimal value. On the other hand, in order to increase the probability of particles jumping out of local optimum, according to the principle of partition crossover, reset the particles that are trapped in local optimum. Next, the multi-window fusion decision result of the adaptive weight voting is used to identify the gesture. Finally, five methods are used to classify the five combined fingers movements about pinch gestures and compare with each other. The experimental results show that the algorithm (IPSO-SVM-WMV) combined sample entropy has strong robustness, real-time and anti-interference ability in double-channel EMGs signal recognition and the classification accuracy is better than PSO-SVM.