Abstract
This paper proposes a method to discriminate between different types of hand and arm motions by applying independent component analysis (ICA) to integrated surface electromyograms (IEMGs). Independent components corresponding to the motions can be derived from the data set of the IEMGs that are measured while a participant performs finite motions sequentially and repeatedly. Then, the obtained unmixing matrix is applied to a new data set of the IEMGs to be classified; the independent components are then calculated. The motion corresponding to the component with the maximum amplitude is determined. The successful derivation of the components corresponding to the respective motions depends on whether the participant can repeat the same motions. A support vector machine (SVM) was also introduced to improve the discrimination performance. Experiments were performed to assess the suggested method, using ten healthy adult participants. As a result, the average correct discrimination rate obtained for each participant ranged from 65.0% to 88.6% when only ICA was applied and ranged from 94.3% to 99.6% after the SVM was introduced. Even if the SVM trained by the data sets obtained from other participants was applied, high correct discrimination rates were obtained.
