Hand Gesture Recognition based on Medium Density Surface Electromyography using Graph Neural Network

Surface Electromyography (sEMG) is a technique used to evaluate and record the electrical activity of skeletal muscles. It provides valuable insights into muscle function and performance, making it a crucial tool in various biomedical applications. Among these, sEMG-based gesture recognition has become a promising approach for human-machine interface in prosthetic devices, supporting amputees in regaining lost motor functions to enhance their daily lives and social interactions.

While recent studies have achieved high gesture recognition accuracy (>90%), these systems still lack robustness in real-world applications. Firstly, sEMG signals are influenced by various non-ideal conditions, such as electrode shifts, muscle fatigue, and changes in arm posture, which are common in everyday settings. Secondly, model performance deteriorates significantly when gesture recognition models are extended to new users, limiting the generalizability of these systems.

The proposed project aims to develop advanced machine learning methods for sEMG-based gesture recognition systems. The candidate will conduct experimental work involving sEMG data collection and machine learning tasks, particularly focusing on graph neural networks and transfer learning. sEMG sensors will be used to gather muscle activity data during specific tasks, and machine learning algorithms will be developed to extract meaningful features from the sEMG signals.

The goal is to enhance the performance of gesture recognition systems across different users and arm positions. The outcomes of this research are expected to have significant applications in prosthetics, enabling individuals with limb impairments to achieve natural and intuitive control over prosthetic devices, thereby improving their quality of life and independence.