Optimized classification of multiclass problems applied to EMG-control of hand prostheses

The article proposes a new control scheme for a multifunctional myoelectric control of hand prostheses. Therefore, switch signals are introduced for movement selection. A finite state automaton changes to corresponding movement states after having analyzed the switch signal. This is done by data processing algorithms like MANOVA, discriminant analysis and maximum-likelihood estimation. However, implementations using recorded data are not able to discriminate all switch signals. Therefore, modifications have been developed to increase classification accuracy, using modified transformation matrices and hierarchical classifiers. These algorithms are tested and compared with data of two above elbow amputees and two below elbow amputees.