Classification of EMG signals by LWRBF network

In this study, a structure with high accuracy for the classification of Electromyographic (EMG) signals is used. This structure is a general-purpose artificial neural network which was proposed in previous studies. This network, called the Linear Weighted Radial Base Function Network (LWRBF) due to the use of a feature extraction strategy which includes Discrete Wavelet Transform (DWT), Principal Component Analysis (PCA) and logarithm function, has a high-accuracy classification ability. The proposed structure, with these high-precision and multi-function properties, can be used in the development of EMG-controlled artificial limbs. We have realized that the used logarithm function with DWT and PCA enhanced a remarkable improvement on the obtained features. EMG data was acquired on forearm muscles as 4 channels for 6 different movements. As a result we have achieved a high classification accuracy rate of 97%.