Conjugate-prior-penalized learning of Gaussian mixture models for multifunction myoelectric hand control

This paper presents a new learning method for Gaussian mixture models (GMMs) to improve their generalization ability. A traditional maximum a posterior (MAP) parameter estimate is used to achieve regularization based on conjugate priors. Plus, a model order selection criterion is derived from Bayesian-Laplace approaches, using the conjugate priors to measure the uncertainty of the estimated parameters. As a result, the proposed learning method avoids the possibility of convergence toward the boundary of the parameter space, and is also capable of selecting the optimal order for a GMM with more enhanced stability than conventional methods using a flat prior. When applying the proposed learning method to construct a GMM classifier for electromyogram (EMG) pattern recognition, the proposed GMM classifier achieves a high generalization ability and outperforms conventional classifiers in terms of recognition accuracy.