Study On Classific Method Of Fingers Actions Vector Based On Semg

The surface electromyography (sEMG) signal, which is noted by the surface electrode, is the electrical manifestations of the neuromuscular activation associated with a contracting muscle. Its application is popular in neuromuscular system to obtain diagnostic information, because of its many virtues such as: collecting signal conveniently and rapidly, non-hurt measurement. By far, many academicians put forward a lot of methods to identify actions including forearm rotating forward or back, wrist extending or bending from sEMG signal. But pattern recognition of fingers'actions is not attached importance to extensively. This paper attempt to study pattern recognition means of fingers'actions based on sEMG..Above all, the physiological theory of sEMG signal and its applications in healing medicine are introduced. At one time, the existing methods of analyzing and identifying sEMG signal are discussed. This knowledge is foundation of next study.Afterwards, Two analytical means and one classifying machine are combined to study pattern recognition of four fingers'actions as follows: five fingers pinching, five fingers spreading, four fingers bending (excluding thumb), and thumb making a away for the centre of the palm. For one analytic means is wavelet transform(WT). This study uses WT to having multiple-dimensioned decomposition and reconstruct -ing with sEMG, in order to extract relevant characteristic values of wavelet coefficients signals according to energy of different frequency segment. Then having singularity value decomposing to reduce dimension, to bring convenience to next recognition. For another analytic means is wave-packet transform(WPT). This study uses WPT to have decomposing with sEMG signal of four fingers'actions, for the sake of extracting characteristic values according to relative wavelet coefficients'energyrates.Finally, some existing classifying machines and its applications are discussed. This study selects BP Neural Network(NN) to classify four fingers'actions according to extracting characteristics on above two means. The results testify that classification accurate rate is more than 90%. Compared with the method of WT combined with BP NN, the method of WPT combined with BP NN is better in some sort.