Surface Electromyography Signal Classification For Human Computer Interaction

Adrian, Kh'ng Kean Chong (2013) Surface Electromyography Signal Classification For Human Computer Interaction. Project Report. UTeM, Melaka, Malaysia. (Submitted)

	PDF (24 Pages) Surface_Electromyography_Signal_Classification_For_Human_Computer_Interaction.pdf - Submitted Version Restricted to Registered users only Download (881kB)
	PDF (Full Text) Surface_Electromyography_Signal_Classification_For_Human_Computer_Interaction.pdf - Submitted Version Restricted to Registered users only Download (1MB)

Abstract

This project is to design a solution to classify surface electromyography (SEMG) signal for human computer interaction. Raw SEMG signals are collected from a total of eight different subjects by using two surface electrodes placed on hand flexor muscles group when performing hand grips. Dynamometer is used when hand grips is performed in order to capture its force level. Hand grips with 12, 25 and 50 percent of maximum voluntary contraction (MVC) are performed on each subject. Feature extraction of Root-Mean-Square (RMS) and Fast Fourier Transform (FFT) are used on the raw SEMG signals and normalized to its MVC. The signals of RMS and FFT are then trained using one hidden layer Feed Forward neural Network with 5, 10, 15, 20 and 25 hidden neuron network to classify its force strength. The networks performance are evaluated using Mean-Square-Error (MSE) to determine how accurate of classification to the actual force. Neural network with 15 hidden neurons gives best classifications with approximately 87% of classification results having MSE less than 5% on overall classification for all hand grips with 12, 25, and 50 percent of MVC.

Actions (login required)

View Item