Continuous classification of myoelectric signals for powered prostheses using gaussian mixture models

Pattern recognition is a key element of myoelectrically controlled prostheses. Improvements in classification accuracy have been achieved using various feature extraction and classification methodologies. In this paper, it is demonstrated that using a simple and direct approach can achieve high classification accuracy, while maintaining a low computational load; important characteristics for a real-time embedded system. An average classification accuracy of 94.06% was achieved for a six class problem, using a single mixture Gaussian mixture model, along with majority vote post-processing.