Design Of Surface EMG Signal Acquisition System And Research Of Movement Pattern Recognition Methods

Surface Electromyography(SEMG) signal is one-dimensional time series signal which isgot from the skin surface, it can reflect the nerve and muscle functional status in the non-injurystate.In recent years, with the development of computers and other technology, domestic andforeign scholars have studied the surface electromyography gradually in depth, making theSEMG not only widely used in clinical medicine, sports medicine and rehabilitation medicine,but also become the ideal control signal for prosthesis limb. The work of this paper is dividedinto two parts: First, completed the design of the surface EMG signal acquisition system, andthen studied the action pattern recognition algorithm.Digital signal processor is a dedicated embedded processor for digital signal processing, ithave the characteristics of high-speed and low power consumption. The EMG signal is a typicalweak signal which has strong background noise, we can not get the optimal results through theanalog circuits, and it will increase the hardware cost and circuit complexity if there are toomuch conditioning circuit. We implement the signal processing algorithms in the digital signalprocessor will not increase the cost, and have the advantages of high flexibility and reliability.This paper designed the EMG amplifier circuit, filter circuit and the acquisition system based onDSP, the amplifier amplify the EMG signal effectively, and the magnification in the72db~74dbadjustable. According to the effective energy range of the EMG signal, the filtering circuitincluding the10Hz high-pass filter,200Hz low-pass filter and50Hz notch filter. The acquisitionsystem using TMS320F2812as the core, AD976as the A/D converter, and the system achievedthe effective acquisition for EMG. In the experiment, in order to identify the arm movements, wehave realized the algorithm in the DSP to process the collected EMG signal, then converted therecognition result to the corresponding control commands to achieve the purpose of controllingthe manipulator.In order to indentify more arm movements, we need to study more complex patternclassification algorithm. This paper used the wavelet packet energy method and the ARparameter model method to extract features from the collected SEMG signal, and then used theBP neural network and Elman neural network to classify the features. Experimental results showthat the study of pattern classification algorithm make the recognition rate of the action between80%~90%, of which AR model is more suitable for EMG feature extraction. The training time ofBP neural network is shorter than the Elman neural network, but the effect of Elman neuralnetwork is better than BP neural network. In specific applications, we need make comprehensiveconsideration of the pros and cons to determine the specific algorithm.