EXPLOITING EMG SIGNALS FOR THE RECOGNITION OF FINGER FLEXIONS USING WAVELET TRANSFORM AND MACHINE LEARNING

Abstract

Electromyography (EMG) is a technique that measures and records electrical activity in response to a nerve’s stimulation of the muscle. EMG signals are biomedical signals that represent electrical currents generated in muscles during their contraction. EMG signals acquired from muscles require advanced methods for detection, decomposition, processing and classification. Various mathematical techniques have received extensive attention and one of the most popular is Wavelet transform. Wavelet transform is a mathematical tool for analyzing data where the signal values vary at different scales, such as in EMG signals, so it is widely used in EMG signal processing systems. This study explored the potential of applying wavelet transform to EMG signals, which were collected using two sensors placed on the forearms of eight subjects performing individual finger flexions. We experimented with various mother wavelets and decomposition levels to determine the most effective combination. After evaluating the results obtained from training models, we selected the Daubechies wavelet (db1) with a second level of decomposition as the optimal solution. To generate meaningful features from the wavelet coefficients, we extracted time-frequency domain features, which were then used as inputs for training and testing machine learning models. We employed five classification algorithms: K-nearest neighbors, Support Vector Machine, Decision Tree, Random Forest, and Extreme Gradient Boosting (XGBoost). By evaluating and comparing the performance of these algorithms, we demonstrated enhanced accuracy and robustness achieved by the combination of wavelet transform and feature extraction in EMG signal analysis.