Multi-Gradient Surface Electromyography (SEMG) Movement Feature Recognition Based on Wavelet Packet Analysis and Support Vector Machine (SVM)

At present, SEMG is used to identify single movement pattern of forearm, however, there are few studies in multi-scale, multi-gradient movement for arm. In this paper, we exploit wavelet packet decomposition to de-noise signal, and de-noised signal is decomposed by wavelet packet to acquire wavelet transform coefficient matrix. Singular values, as input of support vector machine (SVM) classifier, are extracted from this matrix. Meanwhile, use these singular values to train the multi-gradient classifier, constructed by SVM, of arm movement so that to implement the wrist movement such as bend, extend, and rotate wrist slightly or totally, separately; and so that to realize the elbow action such as crook, extend, and rotate arm slightly or totally, separately. The average recognition rate of these movements is above 90%.