Classification of surface EMG signals based on LDA

MA Zhenghua,QIAO Yutao,LI Lei,RONG Hailong

（School of Information Science & Engineering,School of Mathematics & Physics,

Changzhou University,Changzhou 213164,China）

Received:2015-09-08 Revised:2015-11-04 Online:2016-11-25 Published:2016-11-25

Abstract:

We describe a pattern recognition method which is based on a linear discriminant

classifier.We also make a comparison among the combination types of the five feature

parameters,and analyze the two dimension reduction methods of uncorrelated linear

discriminant analysis (ULDA) and principal component analysis (PCA),as well as the impact

of channel number and window length on the classification of EMG signals.Then the LDA

classifier is used to classify the reduced dimension data.Experimental results are as

follows: 1) the accuracy of the combination of the mean square root and the four order AR

coefficients under four channels and eight channels can reach 90% and 96% respectively;2)

increasing the number of channels or features can further improve the accuracy rate;3) when

reducing the dimension of feature vectors to six dimensions by using the ULDA,a high

accuracy rate can still be ensured;4) the recognition accuracy of six hand gestures are

more than 94%,among which the recognition accuracy of four hand gestures can reach more

than 97%.Classification errors mainly occur in the transition phase.

Key words:

MA Zhenghua,QIAO Yutao,LI Lei,RONG Hailong.

Classification of surface EMG signals based on LDA

[J]. Computer Engineering & Science.

PDF