On-line Learning Based Electromyogram to Forearm Motion Classifier with Motor Skill Evaluation.

An on-line learning based EMG to motion classifier can manage learning data set by manual appending and automatic elimination compared with conventional off-line learning based classifiers. It is designed to track the alteration of an operator's characteristics through time. The automatic elimination is based on the continuity of human motion. Moreover, in this study we quantify the attainment of motor skin using the classifier. By classifying up to eight forearm motions from two channels of EMG, we investigate the effectiveness of the automatic elimination process, the validity of the attainment of motor skill by seven trials on an unskilled subject, as well as the relationship among the number of electrodes, the classification performance, and the subject's motor skill. Results show that the proposed approaches can simplify decision boundaries, the attainment of motor skill can be used for judging completion of the training by external observers, and bottlenecks in this classifier can be detected.