A Multi-scale Temporal Convolutional Network-based Method for sEMG Upper Limb Motion Intention Recognition

Surface electromyographys (sEMG) is an electrical signal generated by human muscle contraction, it is easy to obtain, less invasive, and contains rich motor information. It is widely used in medical and rehabilitation fields. In this paper, we propose a motion intention recognition method based on a multi-scale temporal convolutional network(TCN) model. We extract the time domain features and frequency domain features of each channel of sEMG to form a feature vector. Then we use TCN of different sizes to extract features of different scales for feature fusion to improve the recognition accuracy. The experimental results show that the average recognition accuracy is 85.1% for 12 upper limb movements, which is better than the traditional machine learning methods and single-scale TCN, verifying the superiority of the proposed method.