Human Gesture Recognition Using Multifractal Detrended Fluctuation Analysis and Surface Electromyography

Most daily activities need using hands and fingers dexterously. Hand prostheses in disabled people can be controlled using surface Electromyography (sEMG) signals acquired non-invasively by means of surface electrodes connected to superior limbs. After preprocessing 12 electrodes sEMG signals acquired from 10 amputees, different features in time and frequency domains were computed. Considering sEMG as a complex, random, non-stationary, and nonlinear signal a complex nonlinear feature was also extracted by the method of multifractal detrended fluctuation analysis (MFDFA). Different classification methods including support vector machine (SVM), linear discriminant analysis (LDA), and Multi-Layer Perceptron (MLP) were used to compare their performance in the classification of eight different finger movements. It was observed that the SVM performed better than the two other classifiers in finger movement classification. The best classification accuracy, precision, and recall (sensitivity), by the fusion of the new and traditional features were 98.70%, 98.74%, and 98.67%, respectively. Results showed that addition of the new feature extracted by MFDFA and other traditional features was effective in improving the data acquisitions.