The Fifth Main Dynamic Factor: Skin Temperature and Its Effects on EMG-Based Gesture Recognition

Surface electromyography (sEMG) has become a cornerstone for gesture recognition in prosthetic control and human-machine interfaces. However, its performance is often compromised by several dynamic factors, including gesture intensity, limb position, electrode shift, and signal non-stationarities. This work introduces and investigates skin temperature as the "fifth" dynamic factor influencing sEMG-based gesture recognition. Synchronized sEMG, photoplethys-mography (PPG), and inertial measurement unit (IMU) data were recorded from 12 participants without limb differences performing five hand/wrist gestures under three temperature conditions (cold, baseline, and hot) using the BioPoint wearable device. Analysis revealed that cold conditions increased the mean absolute value and reduced the median frequency of sEMG signals, while the signal's complexity (assessed via fuzzy entropy) remained largely unchanged. Gesture recognition models trained exclusively on baseline data showed a decline in accuracy when tested on non-baseline temperature conditions. In contrast, training with temperature-diverse data improved classification robustness across thermal conditions, albeit at the cost of baseline performance. Furthermore, while a multisensor approach combining EMG, PPG, and IMU data enhanced baseline accuracy, it also demonstrated heightened sensitivity to temperature variations. These findings underscore the necessity for skin temperature-aware strategies for the development of robust sEMG-based gesture recognition systems. Additionally, we introduce a novel hand gesture dataset collected under varying skin temperature conditions to support future research on developing more adaptive and reliable gesture recognition solutions.