Research of Motion Classification Based on EMG Signals Pattern Recognition

Electromyogram (EMG) signal is generated by muscle contraction, and surface electromyography signal (sEMG) is recorded by surface electrodes with comfortableness and no trauma, which has been applied in various fields, such as prosthetic control, sports monitoring, rehabilitation technology, and clinical diagnostic. During bio-signal acquisition process, normally cumbersome electronic recording device is used to collect and process the bio-signal, which is apparently not convenient and comfortable for amputees to control the prosthesis device. Furthermore, motion classification algorithm plays a key role in the gesture extraction and prosthetic motion control, which could be further optimized by advanced classifier. This thesis focuses on sEMG signals acquisition using an intelligent wearable wrist-brand and the follow-up data process through bio-signal classifiers. The classification performance comparison has been made between Liner Discriminant Analysis (LDA) and other classifiers. In this study, the sEMG signal was acquired using Myo armband, which a wearable wireless device for gesture control. A data acquisition and processing program was developed to interface the PC based backend processing platform with Myo. The collected sEMG signal were saved for offline analysis, which was latter packed in a specific structure of recSession for feature extraction performed in BioPatRec, an open- source research platform for pattern recognition and prosthesis control. At the end of the data processing flow, in order to compare the performance difference between LDA and other classifiers, corresponding tests were made to verify the performance of a variety of classification algorithms in offline and real-time manner. According to the results, LDA has a higher accuracy, and the fastest training time, which outperforms other classifiers