| A multifunction myoelectric hand controlled by using EMG signals mimics human hand, and is convenient to operate for the handicapped. However, there is a long road to go for its extensive application because the means of identify the hand movement patterns based on the EMGs are immature. Surface EMG signal is a kind of weak physiological signal, and it is of the characteristics of random signal. The current methods of identifying the hand movement patterns based on the EMGs are not so good, there still exist about 20% errors. Any error in identifying the hand movement patterns based on the EMGs will lead serious disaster to the handicapped. So the keys to design a usable myoelectric hand are how to identify the hand movement patterns correctly based on the EMGs. The thesis performs some studies on identifying methods of hand movement patterns based on the EMGs. The main work is shown as follows:1. Preprocessing of the EMG signals: Denoising of the collected EMG signals by wavelet analyze, adopting the layering threshold & the soft threshold method, and carrying it on a comparison with the total threshold method.2. Feature extraction of EMG signals: decomposing the actual signal with wavelet packet, using the standard of the minimum entropy to choose the coefficients of selected frequency range, computing their energy value to form the feature vector.3. Pattern classification: selecting LVQ network as the classifier, comparing its classification performance with BP network's, then pointing out the limitation of the BP network.4. A preliminary study for real-time control of myoelectric hand: propose a scheme to set the length of the moving window, and a novel feature projection method which combines wavelet packet transformation and PCNN, the study lays the foundation for the real-time control of the myoelectric hand.This project is sponsored by the National Natural Science Foundation of China (Project No. 50375108) and the Natural Science Foundation of Tianjin (Project No. 033601611). |
PDF Full Text Request