Dual-Mean Extraction Method of Dynamic Spectrum for Suppressing Random Noise and Coarse Error

Abstract

Dynamic spectrum (DS) can greatly reduce the effects of individual differences and measurement environment by extracting the equivalent absorbance of pulsating arterial blood at multiple wavelengths. DS is expected to achieve non-invasive detection of human blood components. This paper proposes a dual-mean extraction method to reduce random noise and coarse error caused by the jitter of the subject in the acquisition process. The ''superposition averaging'' strategy is adopted twice in this method to reduce the interference of random noise, and a more rigorous standard is used to screen and eliminate coarse error. The single-trial extraction method, the optimizing differential extraction method and the dual-mean extraction method were used to extract the DS from the clinical data of 231 volunteers respectively. The Radial basis function (RBF) neural network was used to model the experimental data of the three extraction results. The difference between the correlation coefficients of calibration set and prediction set was used to evaluate the overfitting degree of the model. The results showed that, the correlation coefficients of prediction set of the data extracted by the dual-mean method reached 0.793, which was higher than the 0.714 of the single-trial method and the 0.749 of the optimizing differential method. And the overfitting degree of the model built by the data of the dual-mean method is the lowest in the next 10 sets of repeated modeling experiments. Compared with the single trial and optimizing differential extraction method, the dual-mean extraction method is better in suppressing random noise and coarse error in DS, and can obtain spectral data with higher signal-to-noise ratio.