STCNet: Spatio-Temporal Cross Network with subject-aware contrastive learning for hand gesture recognition in surface EMG

Jaemo Yang, Doheun Cha, Dong-Gyu Lee, Sangtae Ahn

Abstract: Highlights•This study proposes spatio-temporal cross network (STCNet) and subject-aware contrastive (SAC) loss to mitigate the impact of inter-subject variability while ensuring robust overall performance for hand gesture recognition in surface electromyography (sEMG) data.•The proposed STCNet including rolling convolution effectively captures the geometrical topologies of the sEMG measurement device.•Through the proposed SAC loss learning framework, we align embedding vectors using subject information to further reduce inter-subject variability.•We achieve the state-of-the-art performance on benchmark datasets for hand gesture recognition in sEMG data and validate the proposed methods through extensive ablation studies.