Motion Classification Based on Bispectrum Analysis of Surface EMG Signal

In order to improve the accuracy of upper limb movement recognition, an action classification method based on bispectrum analysis of surface EMG signals was presented in this paper. Information gain was used as the measure criterion of the surface EMG start and end signal segmentation. The segmented signal was extracted by the TKE operator, then the extracted signals were bispectrum-transformed, and bispectrum slices were extracted as the surface EMG features. The probabilistic neural network was used as the classifier, with 100 times 10-fold cross validation as an action classification experiment, and the average correct rate of 10 times was calculated. The correct rates of classification for diagonal slices, secondary diagonal slices and bi-diagonal slices were 94.56%, 90.83% and 95.48% respectively.