Real-Time Monitoring and Injury Risk Warning Algorithm Based on Data Fusion of Motor Muscle Contraction

Abstract

The existing sports injury warning algorithms are difficult to achieve dynamic injury risk modeling based on individual muscle characteristics while ensuring real-time fusion efficiency of multimodal signals. This article proposes a real-time muscle contraction state perception and injury risk warning algorithm based on adaptive weighted feature fusion and lightweight temporal modeling. The system synchronously collected surface electromyography (sEMG), Inertial Measurement Unit (IMU), and pressure distribution data, extracted sEMG time-frequency features through sliding window wavelet packet decomposition, and constructed motion phase context by combining IMU attitude angle and angular velocity. This article further designed a gating fusion module guided by channel attention, dynamically weighting multi-source features and compressing redundant information. On this basis, an individual baseline calibration mechanism was introduced to update the muscle fatigue index and risk threshold using online incremental learning, and a personalized warning model was constructed. The experiment was validated with 12 participants performing high-intensity movements such as squats and sprints. The end-to-end delay of this method is less than 50 ms; the muscle activation recognition F1 score reaches 92.3%; the risk warning lead time is 2.1 s, and the false alarm rate is only 6.4%, significantly better than existing single-modal or static fusion methods. The results indicate that the algorithm effectively improves warning accuracy and individual robustness while ensuring real-time performance, providing a scientific and engineering feasible solution for wearable sports injury protection systems.