LASENet: BiLSTM-Attention-SE Network for High-Precision sEMG-Based Shoulder Joint Angle Prediction

Accurate prediction of shoulder joint angles based on surface electromyography (sEMG) signals is critical in human–machine interaction and rehabilitation engineering. However, due to the shoulder joint’s complex degrees of freedom, dynamically varying muscle coordination patterns, and the susceptibility of sEMG signals to cross-talk and noise interference, achieving high-precision prediction remains challenging. In this study, LASENet (BiLSTM–Attention–SE Network) is proposed as an end-to-end deep learning framework that integrates a bidirectional long short-term memory network (BiLSTM), a multi-head self-attention (MHSA) mechanism, and a squeeze-and-excitation (SE) block to predict shoulder joint angles across three degrees of freedom directly from raw sEMG signals. By jointly modeling temporal dependencies, long-range feature interactions, and channel-wise importance, LASENet effectively captures motion-related patterns while suppressing redundant noise. Experimental results demonstrate that LASENet demonstrates outperforms baseline models in terms of root mean square error (RMSE) and correlation coefficient (CC), achieving superior prediction accuracy and stability. These findings demonstrate that LASENet is an effective solution for accurate shoulder joint angle prediction from sEMG signals.