EMG Signal Based Pattern Recognition of Grasping Movement Using MODWT And Weighted K- Nearest Neighbor

Abstract

This study purposed and evaluates a method based on weighted K-NN classification of surface Electromyogram (sEMG) signals. The sEMG signal classification plays the key role in designing a prosthetic for amputee persons. Wavelet transform is new signal processing technique, which provides better resolution in time and frequency domain simultaneously. Due to these wavelet properties, it can be effectively used in processing the sEMG signal to determine certain amplitude changes at certain frequencies. This paper propose a Maximal overlap Discrete Wavelet Transform (MODWT) approach for Weighted K-NN classifier for classification of sEMG signals based Grasping movements. At level 5 signal decomposition using MODWT, useful resolution component of the sEMG signal is obtained. In this paper Time-domain (TD) features set is used, which shows a decent performance. In WKNN, use a square-inverse weighted technique to improve the performance of the K-NN. Hence, a novel feature set obtained from decomposed signal using MODWT is used to improve the performance of sEMG for classification. MODWT was used for de-noising and time scale feature extraction of sEMG signals. Several WKNN classifiers are tested to optimize classification accuracy and computational problems. PCA is use to reduce the size of the level 5 decomposed data. WKNN performance evaluation on K=10 values with or without PCA. Six hand grasping movements have been classified, results indicate that this method allows the classification of hand pattern recognition with high precision.