Recognition of basketball movement sEMG signals based on multi-channel feature fusion network

This study addresses the underutilisation of multi-channel surface electromyography (sEMG) features in basketball motion recognition by proposing a spatiotemporal fusion network. Multi-channel sEMG signals from athletes' key muscles were collected and synchronised with motion capture data, followed by preprocessing to reduce individual variations. The dual-branch architecture integrates time-frequency feature extraction using convolutional-recurrent networks with graph-based modelling of inter-muscle spatial correlations. An adaptive attention mechanism fuses temporal dynamics and spatial synergies for classification. Experimental results demonstrate superior recognition performance compared to conventional machine learning and single-channel deep learning approaches, with ablation studies confirming the critical roles of spatial modelling and feature fusion. The framework provides an effective solution for analysing complex sports motions through multi-channel physiological signals, offering applications in athletic training optimisation and injury risk prevention.