The real-time recognition of upper limb micro motions based on sEMG signals

Myoelectric interface offers a promising tool for detecting motion intention and extent of movement effort. However, how to achieve motion intention recognition accurately and fast using electromyography (EMG) is an important issue. Many studies present great recognition accuracy, while there is few studies focus on motion recognition speed improvement through exploring motion trend (micro motion) decoding, which is of key importance for the online control strategy of rehabilitation robot. In this paper, we explored the performance of machine learning algorithm in micro motion recognition. The performance of support vector machine (SVM) model was tested for five upper-limb micro motions. As a result, the SVM-based model provides satisfying online performance across all the subjects and motions, achieving an accuracy of 89.7±3.9 % and a total motion recognition time of 0.74±0.08 s. The results show that machine learning algorithm combined with EMG technology can provide accurate and fast upper-limb micro motion intention recognition.