Abstract:

Vast information regarding muscle activity for clinical and engineering applications can be obtained via the EMG (Electromyogram). EMG signal are acquired through surface electrodes which are placed on target muscle set of healthy subjects aged between 23 to 30 years. In this work, six forearm movements have been chosen for classification purpose for both left and right hand. With Hilbert Huang Transform method a total of 21 features of time-frequency domain are extracted for 10 healthy subjects and classified using conjugate gradient method of supervised learning technique using artificial neural networks (ANN). The average accuracy at IMF-1 level obtained is 85.8% for left hand movements, and 86.2% for right hand movement classification. The results of using the Hilbert Huang Transform based ANN classification are quite promising when compared to another classification techniques as K-NN, QDA, LDA and Mahdi Khezri et al. different signal acquisition and classification techniques. The technique can be used for practical implementation of prosthesis for movement classification.Machine Learning algorithms (Decision Trees and Support Vector Machines) are proposed and compared to select a classification system for EMG signals to improve the performance of pattern recognition for the control of a prosthesis prototype. The training, validation and signal classification were made using the Classification Learner application of the MatLab software, using a database captured with the commercial myoelectric armband MYO which contain the information of eight different hand movements. The results show that Support Vector Machines algorithms have a better performance than the decision trees, reaching the 99.8% of accuracy with linear and quadratic kernel and the 99.9% using a cubic kernel

Description:

classification, machine learning, EMG, Artificial Nural Networks, prosthesis