sEMG control of an upper limb rehabilitation robot based on boosting of neural networks

This paper presents a surface electromyography (sEMG) control strategy for robot-assisted upper limb rehabilitation after stroke which can make the rehabilitation robot follow the patient´s intention. A new method for feature extraction is proposed aiming at non-stationary feature of sEMG firstly. And then, an ensemble classification method based on BP base classifier is brought forward to discriminate upper limb motions. Experimental results verify that the feature extraction method is superior to traditional ones with respect to recognition rate and convergence speed of classifier, and the ensemble classifier have stronger generalization ability and higher recognition accuracy than single neural network classifier.