Discrete Wavelet Packet based Elbow Movement classification using Fine Gaussian SVM

Abstract

In this paper; denoising; feature extraction and classification are proposed based on Wavelet Packet Transform(WPT) of Surface Electromyogram signal (sEMG) using Fine Gaussian SVM. sEMG signal was acquired from nine healthy subjects. And then; pre-processing was carried out using MyoResearch XP Software. Further; denoising was improvised using Discrete Wavelet Packet Transform (DWPT). Mother wavelet Daubechies (db 7) at level 4 (Shannon Entropy function) with global threshold (soft) value 1.007 was used for Time-Frequency Domain(TFD) feature extraction so that the possibility of Mean Square Error(MSE) should be minimum. Using Fine Gaussian SVM; an average accuracy of 92.8% and overall MSE of 7.2% was achieved. Five-fold cross valediction methods were performed in this study. The main advantages of this approach are low training time (approximately 2 seconds) and fast speed of response. Using this classification approach; controlling the external robotic device can be achieved in a smooth manner.