Application Of Spectrogram And Discrete Wavelet Transform For EMG Pattern Recognition

• Main Authors: Too, Jing Wei, Abdullah, Abdul Rahim, Mohd Saad, Norhashimah, Mohd Ali, Nursabillilah, Tengku Zawawi, Tengku Nor Shuhada
• Format: Article
• Language: English
• Published: JATIT & LLS 2018
• Subjects: T Technology (General); TK Electrical engineering. Electronics Nuclear engineering
• Online Access: http://eprints.utem.edu.my/id/eprint/23004/2/Application%20of%20Spectrogram%20and%20Discrete%20Wavelet%20Transform%20For%20EMG%20Pattern%20Reocognition.pdf; http://eprints.utem.edu.my/id/eprint/23004/; http://www.jatit.org/volumes/Vol96No10/24Vol96No10.pdf

Electromyography (EMG) pattern recognition has recently drawn the attention of the researchers to its potential as an efficient manner in rehabilitation studies. In this paper, two time-frequency methods, discrete wavelet transform (DWT) and spectrogram are employed to obtain the time and frequency information from the EMG signal. Seventeen hand and wrist movements are recognized from the EMG signals acquired from ten intact subjects and eleven amputee subjects in NinaPro database. The root mean square (RMS) feature is extracted from each reconstructed DWT coefficient. On the other hand, the average energy of spectrogram at each frequency bin is extracted. The principal component analysis (PCA) preprocessing is applied to reduce the dimensionality of feature vectors. Four different classifiers namely Support Vector Machines (SVM), Decision Tree (DT), Linear Discriminate Analysis (LDA) and Naïve Bayes (NB) are used for classification. By applying SVM, DWT achieves the highest mean classification accuracy of 95% (intact subjects) and 71.3% (amputees). To validate our experimental results, the performance of DWT and spectrogram features are compared to other conventional methods. The obtained results obviously evince the superiority of DWT in EMG pattern recognition.