A Study Of The Hand Gesture Recognition And The Classification For A Long Period Of Time Based On Wavelet Neural Network

To improve the living quality of the upper limb amputee,the surface electromyogram(sEMG)signals of residual muscles can be used to control the myoelectric prosthetic hand.Considering the limited residual muscles,it is valuable to investigate the effective pattern recognition algorithms to deal with the sEMG signals detected by fewer sensors,while identifying as many hand gestures as possible.Due to the gradual changes in sEMG characteristics caused by temperature,skin resistance,and other factors,it is difficult to employ the fixed pattern recognition model in identifying hand gestures stably for a long period of time.Therefore,it is necessary to update the pattern recognition model to adapt to the gradual changes in sEMG characteristics.In this thesis,we only used three sEMG sensors to classify and recognize six kinds of hand gestures.The feature extraction methods in time domain,time-frequency domain,and nonlinear dynamics were compared.Then the wavelet transform(WT)was chosen to extract the sEMG feature according to the classification accuracy rate.In addition,we employed wavelet neural network(WNN)as pattern recognition model whose design contains selecting the most suitable mother wavelet activation function.The back propagation and gradient descent algorithms were utilized to train WNN to get the optimal parameters.Six healthy subjects participated in the experiment.The average classification accuracy rate of the proposed WNN is 94.67%,which is substantially better than the results of BP neural network and support vector machine(SVM).Moreover,the experimental results of two upper limb amputees show that WNN still have advantages.It turns out that the fixed pattern recognition model cannot follow the changes in sEMG characteristics for a long period of time.In order to solve this problem,we proposed the incremental learning of WNN ensemble to adapt to the changes in sEMG characteristics.In addition,negative correlation learning(NCL)was used to train the WNN ensemble to get the optimal parameters.Ten healthy subjects executed six hand gestures in a continual experiment for more than two hours.The experimental results demonstrate that our proposed method can recognize the gradual changes in sEMG characteristics stably and improves the recognition effect of the fixed pattern recognition model with an average recognition accuracy of 92.17%.In conclusion,we achieved good results of hand gesture recognition based on sEMG by using WT as the feature extraction method and the WNN algorithm as the pattern recognition model.At the same time,when dealing with the changes in sEMG characteristics,the incremental learning of WNN ensemble is proved feasible with the effective performance of hand gesture recognition.In the future,the proposed method can be used for the real-time control of the myoelectric prosthetic hand for the upper limb amputees.