摘要
Wearable inertial measurement units (IMUs) are widely used in smart sports training, where punch force is an important measurand in boxing but is challenging to measure directly without specialized force sensors. To reduce system complexity and user burden, this work investigates single-sensor-based punch force measurement using only one IMU per hand. We propose an indirect measurement framework that integrates IMU sensing, physics-inspired feature design, and a mixture-of-experts (MoE) model to map the sensor signals to punch force. The application of MoE in sensor data measurement and analysis is rarely addressed in the literature. This work is the first to leverage the MoE model in IMU sensor data analytics for boxing sports. Our method enhances the accuracy of punch force estimation by reducing the root-mean-squared error (RMSE) by 12.27% compared to other methods in the experiments while requiring only a short training time (e.g., less than 30 s).