In recognition of walking gait modes using surface electromyography (sEMG), the use of sEMG sensor array can provide sensor redundancy and less rigorous identification of sEMG electrode placements as compared to the conventional sEMG electrode placements right in the middle of muscle bellies. However, the potentially lesser discriminative and noisier sEMG signals from the sEMG sensor array pose the challenge in developing accurate and robust machine learning classifier for walking activity recognition. In this paper, we explore the use of convolution neural network (CNN) classifier with frequency gradient feature derived from EMG signal spectrogram for detecting different walking activities using an sEMG sensor array on thigh circumference. EMG dataset from five healthy subjects and an amputee for five walking activities namely walking at slow, normal and fast speed, ramp ascending and ramp descending are used to train and test the CNN-based classifier. Our preliminary findings suggest that frequency gradient feature can improve the CNN-based classifier performance for walking activity recognition using EMG sensor array on thigh circumference.
|Title of host publication||International Conference on Intelligent Systems Engineering (ICISE)|
|Number of pages||6|
|Publication status||Acceptance date - 5 Mar 2018|