EMGNet: A large-scale EMG dataset for locomotor intent recognition

Surface electromyography (sEMG) has been used to control bionic legs through locomotor intent classification (i.e., determining locomotor tasks such as incline stairs or level ground). However, current EMG systems have typically been developed using small-scale data with limited individuals and historically have performance limitations between users and sessions due to differences in sensor positioning and/or user physiology. Here, we developed EMGNet to support the development of robust EMG-based intent recognition systems trained and evaluated on large-scale data. The dataset builds on many leading open-source datasets of locomotor tasks published to date – outlined in section 3 – by creating a large meta-dataset with EMG signals processed and prepared for developing intent recognition systems. Each dataset included has been modified to remove unneeded data (such as IMU or other sensors), achieve a consistent data structure for ease of use, and establish a standardized pipeline for preprocessing (filtering, normalization, windowing) and training.