On the Use of Kernel Fisher Discriminant Analysis as a Reduction Method for the Classification of EMG Signals

The muscular system is quite complex and can be divided into smaller systems. In this paper, we focus on the abdominal wall, in particular on the two deepest muscles that compose it, the transversus abdominis and the obliquus internus. Both muscles play an important role in many physiological phenomena, such as breathing. The purpose of this paper is to identify the co-contraction patterns of these two muscles. To this end, we use a combination of two well-known methods. Firstly, the kernel Fischer discriminant analysis (K-FDA) is used to transform the data extracted from surface electromyographic signals, acquired from one bipolar electrode, in order to build a representation of the data that facilitates the classification. Then, a support vector machine is used for the classification step. We tested four types of kernel for the K-FDA, namely linear, radial basis function, sigmoid and polynomial. Following a five-fold cross validation, we obtained an accuracy up to 100%.