Research on sports exercise posture monitoring and energy consumption law based on energy power sensors

This study develops a self-powered multimodal wearable sensing framework for simultaneous sports-posture recognition and human energy-expenditure estimation. The wearable node integrates triaxial acceleration and angular-velocity sensing with an auxiliary micro-energy-harvesting and power-management unit. Sixty participants completed eight representative physical-education movements, producing 2400 subject-independent motion sequences and 720 energy-expenditure bouts. A fourth-order zero-phase Butterworth filter, median filtering, calibration, and time synchronization were applied before classification. A one-dimensional convolutional neural network (CNN) was used to learn temporal representations, and a support vector machine (SVM) performed the final posture classification. The proposed CNN-SVM pipeline achieved 95.4% test accuracy, a macro-F1 score of 95.2%, and a mean inference latency of 6.3 ms. For energy-expenditure prediction, Bayesian optimization with a Gaussian-process surrogate and an expected-improvement acquisition function reduced the root mean square error from 1.12 to 0.82 kcal/min and the mean absolute error from 0.84 to 0.61 kcal/min. The optimized model also outperformed a standard metabolic-equivalent baseline. These results demonstrate that posture context, physiological variables, and motion intensity can be jointly modeled to support low-latency exercise monitoring and individualized energy-management feedback.