Recognizing the Gradual Changes in sEMG Characteristics Based on Incremental Learning of Wavelet Neural Network Ensemble

Most myoelectric prosthetic hands use a fixed pattern recognition model to identify the user's hand motion commands. Since surface electromyogram (sEMG) characteristics vary with time, it is difficult to employ the fixed pattern recognition model in identifying hand motion commands stably for a long period of time. In order to adapt to the gradual changes in sEMG characteristics, we utilized incremental learning based on the wavelet neural network (WNN) ensemble, and used negative correlation learning (NCL) to train it. To verify the effect of the proposed method, a group of subjects executed six hand motions in a continual experiment for more than 2 h. Compared with the fixed pattern recognition model, the classification accuracy rate of incremental learning with nonintegration becomes substantially improved. In addition, the results of the WNN ensemble with the fixed-size mode are more stable than those of the WNN ensemble with the growth mode. The experimental results demonstrate that our method can recognize the gradual changes in sEMG characteristics stably. Using the proposed method, the average accuracy rate is found to be 92.17%, even after a long period of time. Moreover, since the update time is short, the proposed method can be successfully applied in myoelectric prosthetic hands.