sEMG Signal-Based Prosthetic Hand Control Using Embedded Machine Learning with Deep Neural Network

This study focuses on creating a prosthetic arm by leveraging surface electromyography (sEMG) signals for controlling the prosthetic arm. This novel approach holds promise for enhancing the quality of life for amputees and signifies a step forward in the field of prosthetic technology. The key contribution of this research lies in the integration of sEMG data collected from five sensors placed on the forearm muscles, the development of a realtime Deep Neural Network (DNN) model, and the installation of the DNN model on a system-on-chip (SoC) which is Arduino BLE 33 and use of multithreaded operation to recognize different hand gestures in realtime. Various gestures were performed and recorded, forming the basis for training a compact machine learning model using Edge Impulse. The resulting model was optimized for deployment on the Arduino Nano 33 BLE Sense, harnessing its machine-learning capabilities. This model eliminates the need for an internet connection or external computation. The developed prosthetic arm's control mechanism was tested against existing methods, demonstrating its ability to interpret user intentions accurately. The system's responsiveness and gesture recognition accuracy is 97.2% with a processing time of ~ 1 ms and a max latency of 100ms which outperformed traditional approaches, marking a significant advancement in prosthesis control as well as affordable. The study successfully pioneers a specialized prosthetic arm catering to amputees lacking wrist functionality. By harnessing sEMG signals and embedded machine learning with Deep Nural Network (DNN), the developed system offers improved accuracy and responsiveness in interpreting user gestures. This innovation holds promise for enhancing the quality of life for amputees and signifies a step forward in the field of prosthetic technology.