A Novel Method of sEMG Signal Segmentation

In our lower limb rehabilitation system, the surface electromyography (sEMG) signal is adopted as the signal of human-machine interface. In order to be able to control rehabilitation robot in real time, the paper proposes a new and real-time sEMG signal segmentation method P&WND that is accomplished by way of the analysis of peak information and adding non-equidistant window function. The implementation of the method mainly contains four steps. The first step is to remove the signal segments whose derivatives are negative, The second step is to get the reference positions based on the information of signal peak and the repeat number of action test, which is carried out iteratively, The next step is to add window function to the signal positions obtained from previous step, The last step is to calculate the feature in each added window, and finally obtain feature vector. In the end, in order to verify the segmentation result accuracy of the new method and effectiveness for action classifier, two experiments are designed. One is to directly view the segmentation result and evaluate the accuracy by experienced operator using visual inspection. The other experiment is to construct a LS-SVM classifier model so as to observe the signal segmentation method on the result of classification. The experiment result shows that compared with Di Fabio's method that is a traditional signal segmentation method, the new method P&WND greatly improves the accuracy of signal segmentation and the correctness of action classification. And at the same time, it greatly reduces the human's subjectivity and time consumption.