Multiclass Myoelectric Identification of Five Fingers Motion using Artificial Neural Network and Support Vector Machine

The research in Neuro-Prosthetics is gaining more significance and popularity as the advancement in prosthetics control allows amputees to perform even more tasks. Indeed, the improvement of classification accuracy is a challenge in prosthetics control. In this research, a system is developed in order to improve the multiclass classification rate. Two classifiers namely Artificial Neural Network(ANN) and Support Vector Machine(SVM) are trained to recognize five different myoelectric motions of hand fingers. The Electromyography(EMG) signals are acquired using surface electrodes placed on the forearm at specific nodes. The signal conditioning is performed using two stage filtering and amplification followed by digitization process. The final version of EMG signals is correlated in joint time and frequency domain for best feature vectors done via Discrete Wavelet Transform (DWT). The feature vectors are used to train the ANN and SVM. The classification results show an exceptional performance of ANN with classification accuracy of 98.7%. over the SVM, which is 96.7%