sEMG-based hand motion recognition by means of multi-class adaboost algorithm

Human motion is closely related to muscle activities. Numerous researches have proved that it is feasible to predict human motions by using sEMG (surface Electromyography) signals. But due to the low signal-to-noise ratio of sEMG signals, the recognition accuracy of sEMG based motions is still limit...

Ausführliche Beschreibung

Gespeichert in:

Bibliographische Detailangaben
Veröffentlicht in:	2017 IEEE International Conference on Robotics and Biomimetics (ROBIO) S. 1056 - 1061
Hauptverfasser:	Zhou, Shengli, Yin, Kuiying, Liu, Zhengxiong, Fei, Fei, Guo, Jinyi
Format:	Tagungsbericht
Sprache:	Englisch
Veröffentlicht:	IEEE 01.12.2017
Schlagworte:	AdaBoost BP neural network Classification algorithms Feature extraction Neural networks NinaPro databse sEMG Signal processing algorithms Thumb Training Wrist
Online-Zugang:	Volltext
Tags:	Tag hinzufügen Keine Tags, Fügen Sie den ersten Tag hinzu!

Beschreibung
Zusammenfassung:	Human motion is closely related to muscle activities. Numerous researches have proved that it is feasible to predict human motions by using sEMG (surface Electromyography) signals. But due to the low signal-to-noise ratio of sEMG signals, the recognition accuracy of sEMG based motions is still limited, especially when there are multiple categories. In this paper, a multi-class AdaBoost based algorithm is introduced to combine the prediction of ten component BP neural networks. From the results of recognizing 8 wrist motions and 12 finger motions, it reveals that the proposed algorithm has the advantage of enhancing the classifiers' performance by weighting them together without bring a burden for real-time application. The mean recognition accuracy of 8 wrist motions over 27 subjects reaches 94.35%±4.11%, while the mean recognition rate of 12 finger motions reaches 93.58%±3.44%, which are much higher than the reported accuracies with the same dataset.
DOI:	10.1109/ROBIO.2017.8324557