Classification of EMG Signals for Myoelectric Prosthesis with Feature Images

Accurate hand movement recognition using electromyography (EMG) data is crucial for creating a reliable human-machine interface system for prosthesis control. Proper feature selection from EMG signals is required for the precise recognition and classification of various movements. In this regard, we proposed a classification scheme that improves classi- fication accuracy. This work focuses on the classification of EMG signals for hand prostheses by using sEMG feature images which are further fed into a simple CNN network. Some commonly used time-domain features such as Root Mean Square(RMS), Mean Absolute Variance(MAV), integral Absolute Variance(IAV), Slope Sign Changes(SSC), and Waveform Length(WL) are utilized in this research. The features are converted into images, further fed into the CNN network for classification. We utilized hand gestures of the benchmark datasets NinaPro DB1 and obtained a classification accuracy of 91.2%.