Classifying sEMG-based hand movements by means of principal component analysis

In order to improve surface electromyography (sEMG) based control of hand prosthesis, we applied Principal Component Analysis (PCA) for feature extraction. The sEMG data (downloaded from free NINAPRO database) were recorded during three grasping and 11 finger movements. We tested the accuracy of a simple piecewise quadratic classifier for two sets of features derived from PCA. Preliminary results from a group of healthy subjects suggest that the first two principal components aren't always sufficient for successful hand movement classification. The grasping movement classification error when using three features (22.7±10.7%) was smaller than the classification error for two features (33.4±12.5%) in all subjects.