In the field of oximetry monitoring research, non-contact physiological parameter detection techniques have been increasingly emphasized for their convenience and non-invasiveness. Conventional methods mainly rely on spatiotemporal images of the face or remote photoplethysmography (rPPG) signals for oximetry estimation, which have been validated for their effectiveness but still have potential for improvement. The aim of this study was to explore the potential link between direct current (DC) trend changes in rPPG signals and blood oxygen saturation, with a view to developing more accurate and stable non-contact oximetry methods. We found a significant correlation between DC trend changes in different channels of the rPPG signal and blood oxygen saturation. This finding provides an important theoretical basis for constructing a novel blood oxygenation estimation model. Accordingly, we propose an innovative Dual Component Decoupling network with rPPG, which captures the amplitude and DC signal variation characteristics among different channels through multi-dimensional signal inputs, and greatly reduces the errors of existing detection methods. Extensive experiments on two public datasets as well as one Self-built dataset show that our method can achieve superior performance over competitors.