Forearm EMG Signals Classification using an Adaptative Neuro-Fuzzy Inference System

An electromyogram (EMG) signal detects the electrical potentials generated within muscle cells. These cells are activated by electrochemical and neural impulses. It is quite challenging to differentiate between the individual waveforms produced by the muscle. Thus, EMG signal classification and analysis have become extremely important. Existing quantitative analysis approaches have several limitations, including a decreased recognition rate of MUP waveforms, sensitivity to continual training, reduced accuracy, and inconsistent output. The current paper focused on the EMG signal classification of four gesture classes as a preliminary study toward prosthetic hand design and development. The Myoware muscle sensor was used to capture EMG readings from four forearm gestures. Following data capture, EMG signal characteristics were retrieved using a wavelet transform. Next, ANFIS-based learning is utilized to classify the retrieved features. The result of the ANFIS classification is 96.5 %, which is higher than the LDA classification that has been conducted in previous studies. By referring to the accuracy value that was achieved, it is clear that the findings of this preliminary investigation can be utilized as input in the prosthetic hand control system that will be built in subsequent studies.