Experimental Study and Characterization of SEMG Signals for Upper Limbs

Surface electromyogram (SEMG) is used to measure the activity of superficial muscles and is an essential tool to carry out biomechanical assessments required for prosthetic design. Many previous attempts suggest that, electromyogram (EMG) signals have random nature. Here, dual channel evaluation of EMG signals acquired from the amputed subjects using computational techniques for classification of arm motion are presented. After recording data from four predefined upper arm motions, interpretation of signal was done for six statistical features. The signals are classified by the neural network (NN) and then interpretation was done using statistical technique to extract the effectiveness of recorded signals. The network performances are analyzed by considering the number of input features, hidden layer, learning algorithm and mean square error. From the results, it is observed that there exists calculative difference in amplitude gain across different motions and have great potential to classify arm motions. The outcome indicates that NN algorithm performs significantly better than other algorithms with classification accuracy (CA) of 96.40%. Analysis of variance technique presents the results to validate the effectiveness of recorded data to discriminate SEMG signals. Results are of significant thrust in identifying the operations that can be implemented for classifying upper limb movements suitable for prostheses design.