SEMG Signal Action Recognition Of Intelligent Wheelchair Man-machine Interface

In recent years, with the present computer technology, neurophysiology, digital signal processing technology and pattern recognition and artificial intelligence techniques such as the rapid development of domestic and foreign researchers for further research into the multi-channel semg. Studies have shown that the multi-channel semg intelligent artificial limbs, intelligent wheelchair, rehabilitation medicine, biological engineering, and other fields have a wide range of applications. In this paper on the basis of study the theory of distributed multi-channel semg, combining with the characteristics of the multi-channel semg, the main research of human upper limb arm to pick up and multi-channel semg multiple action pattern recognition algorithm, and then implement the wrist exhibition boxing, fisting, varus and valgus four action pattern classification. Through this research can provide the intelligent wheelchair based on electromyography with certain reference and practice platform. The main research work and the innovation are as follows:(1) Electromyographic signal acquisition based on labview platform, is divided into acquisition needs multi-channel semg hardware circuit and software system based on labview. Design of the platform can real-time acquisition and processing effective methods of electrical signals.(2)For non-stationary and nonlinear characteristics of the multi-channel semg and multi-scale decomposition and wavelet packet coefficient after the problem of high dimension, in order to characterize effective electromyographic signal at the same time, the greatest degree of reduce the dimension of feature space and simplify the structure of classifier is proposed a multi-scale decomposition of wavelet packet feature representation and pattern recognition methods. To collected the radial muscle, called lateral brachial wrist flexor, feet side wrist flexor and extensor four-way wavelet packet decomposition, multi-channel semg after decomposition of wavelet coefficients. Then, on the one hand, according to the multi-scale decomposition of wavelet packet coefficient and energy multi-channel semg to reconstruct the feature vector, the inner link between, on the other hand, according to the orthogonal wavelet base vector reconstruction of Bao Ji characteristic vector, respectively action pattern characteristic matrix; With time domain analysis and frequency domain analysis and feature extraction methods such as comparative experiments. By using nonlinear regression neural network classifier contrast experiment shows that proposed by the multi-scale decomposition of wavelet packet reconstruction after the eigenvector method is superior to the time domain analysis method and the commonly used frequency domain analysis method, can better characterization and simplify the structure of classifier of semg.(3) In order to improve the efficiency of forearm movement pattern recognition, using the nonlinear regression neural network to surface signal pattern recognition. With bayesian algorithm, Fisher algorithm, BP network, K neighbor, and the SVM(support vector machine) classifier contrast experiment shows that the nonlinear regression neural network classifier characteristics of action pattern recognition is more efficient.Experiments show that, using semg pattern feature representation and recognition method can successfully identify four kinds of motions such as hand grasping, hand opening, radial flexion and ulnar flexion. And action average recognition rate is up to 95.0%.