Study On Feature Analysis Methods Of Upper Limb S Emg Signal

The surface electromyography signal(s EMG) can be monitored noninvasive by using electrodes on the skin surface and it provides an easy access to studying in clinical medicine, rehabilitation medicine, sports medicine, neurophysiology, and etc. With the development of signal processing technology, how to extract the effective and accurate information from s EMG and make the action recognized has become an important research content in human body muscle electric control and rehabilitation for the disabled. The paper mainly aims at the signal de-noising, feature extraction and pattern recognition.In this paper, the experimental platform of a collection of surface EMG signal is constructed. It is based on multiple electrode of s EMG and the system of the MATLAB/XPC Target. The system is a real-time data acquisition by using host and target machine. Six different kinds of action’s s EMG signals of the upper arm have been acquired effectively. The feature analysis methods of upper limb s EMG signal have been further researched. The main work in this paper is as follows:1) The wavelet packet transform has been put forward in the paper. The experiment results show that the wavelet packet de-noising method has the highest SNR, the minimum root mean square error and smoothness. And it can effectively remain a large number of useful signals, and has the best de-noising effect. It can be suitable for the de-noising processing of multi-channel s EMG.2) In this paper, several methods for the feature are researched, such as: time domain, frequency domain and time-frequency domain(the wavelet analysis). Based on the analysis comparison, the results show that the time domain and frequency domain characteristics can not reflect the signal characteristic effectively, but that the maximum value and singular value of the wavelet coefficients can made it, so they could be used as the input vector to construct feature classifier.3) The neural network classification is used in this paper. The momentum BP method and LM algorithm are also used to improve the network. The improved network is compared with the standard BP neural network. And the experimental results show that the recognition rate of the standard BP neural network is the lowest. As a comparison, the momentum BP neural network has been greatly improved in recognition rate. The LM network has achieved the best effect, the convergence speed is the fastest and the recognition rate is above 95%.