Recognizing Hand Motions Based on Fault-tolerant Classification with EMG Signals

In view of the fault/missing data problem caused by disconnected/damaged electrodes and data-transmission interrupting in myoelectric-interface systems, an EMG (electromyography) fault-tolerant classification method based on Gaussian mixture model is proposed, with which an incomplete-data sample can be classified via marginalizing or conditional-mean imputation of the fault/missing data in the EMG feature sample. The proposed method is applied to recognizing five kinds of hand motion. Experimental results show that the proposed method can provide higher motion-recognition accuracy than that by the traditional zero and mean imputation methods. Finally, a myoelectric-hand platform is developed by involving the fault-tolerant classification mechanism, and the online experiments show that the proposed method can effectively improve the robustness of myoelectric-interface systems.