Development of a grip force recognition system based on EMG signals and neural networks

The object of research is a bionic prosthesis control system that uses EMG signals read using the MYO bracelet, as well as feedback sensors to determine the grip force. In the context of the development of modern bioengineering and neurotechnology, this system is aimed at ensuring accurate and adaptive control of the prosthetic hand, taking into account the user's intentions. The problem considered in the research is to recognize the grip force of a bionic hand based on EMG signals and transmit feedback to the user. Special attention is paid to the use of a deep neural network for classifying force levels and developing a real-time signal processing technique. The task is to create a stable and user-friendly grip control system. The essence of the results obtained is to create an experimental system that classifies the grip force of objects with a bionic hand with high accuracy (95%). The system is based on a neural network with a two-layer autoencoder, trained on labeled and unlabeled data. To improve the accuracy of the model, the temporal characteristics of EMG signals were used: MAV, RMS, SD and WL. The results are explained by effective biosignal processing and machine learning. The division of force into 8 levels and the use of a fuzzy controller ensured stable control of the grip and the transfer of information to the user via vibration feedback. The system was successfully tested in real time. The innovation lies in the integration of the MYO bracelet, force sensor and FSR with deep learning. This provides accurate force classification and natural feedback, which increases controllability and ease of use. The use of the system provides new opportunities in prosthetics: it more accurately conveys the user's intentions, reduces errors and increases comfort. The results have the potential for clinical implementation to improve modern prostheses.