Human motion heart rate measurement with MSSF and feature interaction

Conventional heart rate monitoring often struggles in dynamic exercise settings, facing challenges with accuracy and personalized adaptation. To address this, this study proposes a novel model utilizing Multi-Scale Spectral Feature selection and an ensemble of Extremely Randomized Trees and a Multilayer Perceptron. Transfer learning is applied to adapt the model to individual physiological differences. The proposed method achieves 98.7% accuracy in real-world applications and demonstrates high user satisfaction. This approach significantly improves the precision and personalization of heart rate monitoring during complex motion, offering a robust solution for dynamic environments.