Research on Optimized Methods for Dynamic Characteristic Retention and Artifact Removal in Blood Oxygen Signals

This paper presents a hybrid filtering technique combining ensemble empirical modal decomposition (EEMD) and Adaptive Kalman Filter (AKF) based on the new interest variance is proposed to realize dynamic filtering in a nonlinear environment. EEMD is used to decompose the original signals and extract the low-frequency components in order to obtain the main physiological signal features. Meanwhile, AKF dynamically adjusts the filtering parameters to optimize the performance of the algorithm through the new resting variance method. The experimental results demonstrate that in comparison with conventional Kalman Filtering (KF), the proposed method enhances the average signal-to-noise ratio from 1.80 dB to 4.38 dB, reducing the mean square error by 48.95%, and exhibits an average processing time of 1.2 mS. These improvements make the technique particularly advantageous for the processing of pulse oximetry signals with motion artifacts. By balancing noise reduction and information fidelity with low power consumption, this approach is well-suited for wearable devices, offering reliable data for daily health management and disease prevention.