The electromyogram (EMG) signal is an electrophysiological signal that measures currents produced by muscles throughout the human body during contraction, thus representing neuromuscular activity. A common early approach to measuring EMG signals for use as a control input was to place electrodes precisely over specific muscles, now known as sparse multichannel surface EMG. However, to facilitate more general wearable interfaces that could be used in everyday contexts, EMG-based systems must be simple and non-invasive, such as embedded in a socket, a watch, an armband, jewelry, or concealed beneath clothing. Consequently, it is now common to position EMG sensors radially around the circumference of a flexible band. Due to recent advancements and the availability of such EMG sensors, together with advances in wireless communication and embedded computing technologies, EMG data can indeed now be obtained unintrusively using wearable devices. Moreover, impressive advancements have been made in EMG signal processing and pattern recognition over the past few decades. This has greatly increased the number of potential applications for the use of EMG, including, but not limited to, powered prostheses and orthoses, electric power wheelchairs, human–computer interactions, and diagnoses in clinical applications.
Although performance of myoelectric control systems, or EMG pattern recognition, exceeds 90% in controlled environments, myoelectric devices still face challenges in robustness to variability introduced during daily living conditions. Current challenges are commonly associated with this lack of reliability in practical conditions and can be roughly categorized into confounding factors such as limb position, contraction intensity, time (within-day and between-day variability), electrode shift, muscle fatigue, noise, hand-busy conditions, cross-user classification model, etc. New and advanced signal processing and machine learning methods have thus been proposed to minimize the degradation caused by the variation introduced by these aforementioned factors. Robust feature extraction methods, new training strategies, transfer learning and deep learning approaches, and sensor fusion are just some of the emerging and state-of-the-art approaches.
The aim of this Special Issue is to bring together researchers active in the development of EMG sensors, their interpretation, and their applications. Works on innovative EMG signal processing and machine learning algorithms aiming to address critical issues are welcome and encouraged.
Dr. Erik Scheme
Dr. Angkoon Phinyomark