Pattern Recognition Of SEMG Based On Prosthesis

Surface electromyography (SEMG) signal is the spotting signal which is produced when energy is coming from muscle. SEMG is obviously different for the different kinds of motion modes of human arms. Accordingly, the motion modes can be recognized by pattern recognition of SEMG. As for the controlling human-robotic arms by SEMG, it is the essence to the characters extraction and modes category of SEMG. In this paper, six motion modes- are acquired from human case by SEMG signal acquisition apparatus, and their features are also analyzed within the time domain, frequency domain and time-frequency domain. Most importantly, a brand-new method of extraction of feature parameters based on the Fussy Logic is brought out to bi-lateral energy ratio and energy subtraction. Then, first channel square root of energy, second channel square root of energy, and their bi-lateral energy ratio and energy subtraction, all these four characteristics were used as EMG signal's parameters to be classified by applying fuzzy math classifier and decision tree classifier. The classifier of fuzzy math is different from traditional classifier which was based on fuzzy logic. The algorithm do not make data fuzzy. First, all six motions'data were collected to be one set, then the set was cut to become N regions, the distribution ratio of each motion in every region was calculated as each motion's membership in the region. Finally, each motion's product of four memberships was used as criterion. The motion whose product is the biggest one, is the output of the classifier. The average accuracy of this classifier is 85.67%. It has several advantages, such as simple structure, easy to calculate. But, it is not good to classify small amount data. fuzzy cluster which is based on algorithm of fuzzy c-mean is used as classifier to classify semg signal. The average accuracy can reach 91.33% when first channel square root of energy and second channel square root of energy were applied as characteristics. To compare with classifier of fuzzy math, its average accuracy increase so much, but it takes longer time to train data, and it is more complex, some motion's accuracy is very low as well. All these will effect its application in reality. The classifier of decision tree has two types. One is one versus one, the other is one versus many. The one versus one is based on man's thinking in differentiate two different things, that is only two choices"yes or no". If six motions can be seen as one set was composed by two different sets, according to this way, one set can be classified to two sets, finally can classify six motions. The accuracy of this classifier reaches 81%. Its advantage is really like man's thinking logic, but the algorithm is complex, and it's bad to classify the data which were mixed. To solve these problems happened to one versus one classifier, one versus many classifier was made. It can classify all six motions by least step. So it is simple to calculate, and it is efficient to classify mixed data. Its accuracy can reach 92.67%. It is the best one classifier in the text. To compare with Bp neuron network, although accuracy of Bp network is 2.33% higher than one versus many classifier, its calculation is complex and the time of selecting suitable parameters is so long. If to considerate of efficiency, one versus many classifier is better than Bp network.