EMG signal patterns recognition based on feedforward Artificial Neural Network applied to robotic prosthesis myoelectric control

The present work is part of the “Hand of Hope” project, which seeks to develop low-cost robotic prostheses with the aim of contributing to the social and labor inclusion of people with motor disabilities of their upper extremities. The specific objective to address in this work is the design and development of the system architecture to recognition of EMG (Electromyography) signal patterns, the purpose is to control the functions of objects handling by the robotic prosthesis. There are proposed methods to the recognition of EMG signal patterns; however it defined using a Feedforward-backpropagation Artificial Neural Network (ANN) because have high success rate (Sr) using the least amount of channels, can be supported on platforms on-line and real-time; and moreover in order to improve the ANN-Sr values, it was used as input the envelope of the EMG signal instead of the original signal. For the performance evaluation 20 assessments for each of the four EMG signal patterns (relaxed hand/cylindrical grip, pinch grip, thumb adduction, and index finger extended) were performed, from the results it is observed that the average success rate is equal to 95%.