Research On EMG Signal Data Acquisition And Pattern Recognition

This paper researches on the "Upper Stump EMG Testing And Training System",which including amputee's stump surface EMG data acquisition system hardware and software design,and the preprocessing, analysis and feature extraction of the collected data, as well as the pattern recognition based on the EMG characteristics.Firstly, uses S3C44B0X to control the AD sampling and USB data transmission of EMG, and uses C++ Builder to complete the coding of visualization software to design the EMG signal acquisition system.Secondly, preprocesses and analyzes the histogram, covariance, spectrum, sum and difference of the two-way data getting from the amputees when they do different actions, as well as the movements of the beginning and the end of each practice. The results of pre-Analysis show that the data histogram of switch is more compact than the flexion and extension movements in the distribution, and the data of single flexion and extension motion or switch shows weaker correlation as well as time delay increasing. However, the data of continuous flexion and extension motion or switch shows strong Cycle-related correlation. The spectrum of the upper srump EMG envelope is concentrated in less than 150Hz.And then, calculates AR parameters of the sampling data based on Blackman window of the power spectrum ratio and Berg Recursive Methods respectively. Based on the power spectrum ratio and AR parameters, identifies trainers actions through three-layer BP neural network and Support Vector Machine. The results show that Support Vector Machine performance is better than BP neural network algorithm for the upper stump EMG envelope signal characteristics.