Classification of ankle joint movements based on surface electromyography signals

Electromyography (EMG) signal has valuable information about the force of the muscle contraction and the movement direction. This crucial information has been used for many years in exoskeleton, orthoses and prostheses robots. An essential part of those devices is EMG based control system that employs the EMG signal from different muscles to control prostheses and exoskeleton robot. However, using EMG signal as an input control signal for those devices is not easy due to the complexity nature of this signal that produces the different body movements. This difficulty can be overcome by using pattern recognition techniques to discriminant different limb movement’s pattern then use the classified signal as input control signal to manipulate and drive the assistive robot devices. Though much research have been carried out to classify the upper and lower limbs movement based on the EMG signal, still there is a strong need to obtain an accurate pattern classification system in computationally efficient manner. In this work two parts are primarily presented. The first partt was design and implements a multichannel EMG acquisition system to detect and acquire the leg muscles’ signal. In this part four EMG channels were implemented using instrumentation amplifier (INA114) for pre-amplification stage then the amplified signal was filtered using band pass filter to eliminate the unwanted signals. Operational amplifier (OPA2604) was involved for the main amplification stage to get the output signal in volts. The EMG signals were detected during movement of the ankle joint of a healthy subjects. Then the signal sampled at rate of 2 kHz using NI 6009 DAQ card and LabVIEW software was employed to store and display the acquired signal. Fast Fourier Transform (FFT) and Signal to Noise Ratio (SNR) were applied to assess the recoded electromyography signal. The second part is to classify four ankle joint movements which are dorsiflexion, plantar flexion, adduction and abduction. The data was collected from twenty healthy subjects using the custom multichannel EMG acquisition system designed in the first part of this project. In this section, new time domain feature set was evaluated and compared with well known time domain features. Three classifiers were employed to evaluate the two feature sets. These classifiers are linear discriminant Analysis (LDA), K nearest neighbourhood (k-NN) and Naïve Bayes classifier (NB). The result showed the superiority of the new time domain feature set which are the logarithmic based time domain features upon the conventional time domain feature. In addition, the results show the outperformance of LDA classifier among the other two classifiers used in this study.

Text FK 2015 23.pdf - Submitted Version Download (1MB)

Actions (login required)

View Item