MyoBit: A Public Dataset Based on An Armband with 16 sEMG Channels for Gesture Recognition under Non-ideal Conditions

The robustness of surface electromyography (sEMG)-based gesture recognition in practical applications has received much attention due to the influence of external non-ideal factors. Unlike most existing sEMG-based gesture recognition datasets that use sparse or high-density resolution instruments for data acquisition under ideal conditions, this paper proposes a sEMG armband (Biofrontier) with semi-dense resolution that records 7 gestures from 24 subjects (12 male, 12 female) under 9 non-ideal conditions as a public dataset, MyoBit. The results demonstrate that Biofrontier has a high signal-to-noise ratio and repeatability, and the MyoBit is able to achieve a high accuracy of gesture recognition by classical classifiers. Furthermore, two methods for dataset augmentation, increasing resolution and expanding rotation data, have been proposed for researchers. The dataset link: www.biofrontier.cn/dataset.html