Classifying Motion Intention from EMG signal: A k-NN Approach

The use of robotic systems has been investigated over the past couple of decades in improving rehabilitation training of hemiplegic patients. In an ideal situation, the system should be able to detect the intention of the subject and assist them as needed in performing certain training tasks. In this study, we leverage on the information from the electromyogram (EMG) signals, to detect the subject's intentions in generating motion commands for a robotic assisted upper limb rehabilitation system. As EMG signals are known for its very low amplitude apart from its susceptibility to noise, hence, signal processing is mandatory, and this step is non-trivial for feature extraction. The EMG signals are recorded from ten healthy subjects' bicep muscles, who are required to provide a voluntary movement of the elbow's flexion and extension along the sagittal plane. The signals are filtered by a fifth-order Butterworth filter. Several features were extracted from the filtered signals namely waveform length, mean absolute value, root mean square and standard deviation. Two different classifiers viz. Support Vector Machine (SVM) and k-Nearest Neighbour (k-NN) were investigated on its efficacy in accurately classifying the pre-intention and intention classes based on the selected features, and it was observed from this investigation that the k-NN classifier yielded a better classification with a classification accuracy of 96.4 %.