Lower limb motion recognition based on surface electromyography decoding using S-transform energy concentration

Abstract Lower limb motion recognition of based surface electromyography (EMG) aims to provide a more natural and effective human-computer interaction for intelligent prostheses. Accurate motion recognition relies on high-quality EMG decoding, and the key to improving the efficiency of EMG pattern recognition is to optimize signal feature extraction. The phase information of the EMG signal cannot be neglected for motion pattern recognition. Therefore, we proposed a decoding scheme for surface EMG signals based on S-transform energy concentration for lower limb motion recognition, including level walk, stair ascent, stair descent, and crossing obstacles. First, six-channel lower limb EMG signals of the 10 subjects during four kinds of movements were experimentally acquired, and the correlation of multi-channel EMG signals was analyzed to find the best combination of the EMG signal for exploring the classification effect based on the support vector machine (SVM) between single-channel signals and multi-channel signals fusion. The results showed that based on the simple time-frequency domain features with better motion recognition are the semi-tendon muscle and the rectus femoris muscle, and based on the S-transform energy concentration with better motion recognition is the medial gastrocnemius muscle and the rectus femoris muscle. Finally, taking the rectus femoris signal as an example, the motion recognition accuracy of 10 subjects under the two schemes was calculated. The mean value of motion recognition accuracy based on simple time-frequency domain features was 80.71%, and the mean value of motion recognition accuracy based on S-transformed energy concentration was 93.70%. It validated that the S-transform energy concentration scheme has a better recognition effect, and the accuracy of multi-channel signal fusion pattern recognition based on S-transform energy concentration is higher than 96%, which is beneficial to promote the practical application of EMG signals in motion recognition. It has the potential to improve the adaptive human-robot interaction control of the prosthesis.