《A Supervised Feature-Projection-Based Real-Time》.pdf

282 IEEE/ASME TRANSACTIONS ON MECHATRONICS, VOL. 12, NO. 3, JUNE 2007 A Supervised Feature-Projection-Based Real-Time EMG Pattern Recognition for Multifunction Myoelectric Hand Control Jun-Uk Chu, Member, IEEE, Inhyuk Moon, Member, IEEE, Yun-Jung Lee, Member, IEEE, Shin-Ki Kim, and Mu-Seong Mun Abstract—Electromyographic (EMG) pattern recognition is es- a few control commands, such as the opening and grasping of sential for the control of a multifunction myoelectric hand. The the fingers, or the pronation and supination of the wrist. In re- main goal of this study was to develop an efficient feature- cent years, several multifunction myoelectric hands have been projection method for EMG pattern recognition. To this end, a linear supervised feature projection is proposed that utilizes a lin- developed with a number of DOFs and dexterous hand func- ear discriminant analysis (LDA). First, a wavelet packet transform tions [3]–[5]. Yet, such multifunction myoelectric hands require (WPT) is performed to extract a feature vector from four-channel a robust and computationally efficient EMG pattern-recognition EMG signals. To dimensionally reduce and cluster the WPT fea- method to classify and control the hand functions. Nonetheless, tures, an LDA, then, incorporates class information into the learn- an EMG signal is essentially a one-dimensional pattern with ing procedure, and identifies a linear matrix to maximize the class separability for the projected features. Finally, a multilayer percep- a large variation and nonstationary properties. In addition, a tron classifies the LDA-reduced features into nine hand motions. To