Abstract:
Two channel mechanomyographic signals were collected from anterior tibial muscle and peroneus brevis in the subjects’ calves when their ankles acted dorsiflexion, plantar flexion, adduction and abduction, i.e., four action modes. The collected signals were segmented by unequal division algorithm based on signals’ second envelops and nonlinearly scaled wavelets were used to get singular value decomposed (SVD) features from segmented signals, The SVM classifier was used to classify the action modes. Results showed that the second envelop based unequal division algorithm can effectively intercept action segments from original signals and SVD features extracted from two channel signals by nonlinearly scaled wavelets can provide action modes recognition with the overall accuracy rate of 87.8%, which verified the feasibility of the proposed methods.