Surface electromyography–based hand movement recognition using the Gaussian mixture model, multilayer perceptron, and AdaBoost method

Human movement is closely linked with muscle activities. Research has indicated that predicting human movements with surface electromyography signals is feasible. However, the classification accuracy of surface electromyography signal–based movements is still limited due to the low signal to noise ratio, especially when multiple movement categories are investigated. In this study, six representative time-domain feature extraction techniques and four frequency-domain feature extraction techniques with three different types of classifiers (the statistical classifier Gaussian mixture model, the neural network classifier multilayer perceptron, and the ensemble method AdaBoost) were applied for the recognition of 52 movements in Non-Invasive Adaptive Prosthetics database 1. From the experimental results, we observed that the performance of Gaussian mixture model was superior to that of the multilayer perceptron in both classification accuracy and computational load. When AdaBoost was introduced into the multilayer perceptron, the classification accuracy significantly improved, such that the performance was comparable with that of the Gaussian mixture model. Using the combination of the Gaussian mixture model and the mean of absolute value, we achieved an accuracy rate of 89.5% for the classification of the 52 movements, which was much higher than the 76% rate reported in previous studies.