Real-time Oxygen Uptake Estimation in Hypoxic Environments Based on Enhanced GRU Network with Domain Adaption

Real-time oxygen uptake is a crucial indicator in the field of high-altitude medicine. With the development of tiny wearable sensors and machine learning, oxygen uptake estimation via multi-physiological parameter fusion has drawn attention. However, existing methods mainly target the plain environment. Considering the significant impact of high-altitude hypoxia on oxygen uptake and the difficulty of collecting hypoxic data, we propose an enhanced Gate Recurrent Unit (GRU) network with domain adaptation. The proposed method is capable of constructing the general oxygen uptake estimation model for different environments by only using a few data from the target hypoxic environment. Extensive experiments demonstrate that, the proposed method surpasses the state-of-the-art data-driven methods in terms of prediction accuracy and stability. The MAE (Mean Absolute Error) of GRU - DA under the four exercise states of rest, walking, climbing, and running are 1.79, 2.64, 3.2, and 3.95 ml/(kg·min) respectively. The comprehensive error is 3.14 ml/(kg·min), representing a 29.28% improvement compared to the original model.