A multi-band spectral data fusion method for improving the accuracy of quantitative spectral analysis.

The signal-to-noise ratio of the spectrum is a critical determinant of detection accuracy in compositional analysis utilizing spectroscopy. The spectral data acquired by the spectrometer, while intended to capture essential sample characteristics, is often interspersed with various noise interferences. This contamination can severely disrupt the integrity of measurement outcomes. Therefore, this paper proposes the "multi-band spectral data fusion" method. In order to verify the feasibility of this method, this paper takes blood detection based on dynamic spectroscopy as an example and develops two models for each of the various components in blood. The experimental results show that when compared to modeling the raw spectrum data of the samples directly, the prediction accuracy of the model constructed using the new spectra processed by the multi-band spectral data fusion method suggested in this paper is greater. The correlation coefficient of the hemoglobin prediction set has improved by 13.48 %, and the root mean square error has decreased by 21.00 %. The correlation coefficient of the blood glucose prediction set improved by 4.07 %, and the root mean square error decreased by 12.78 %. The result demonstrates that the proposed method effectively mitigates the impact of random errors without compromising the spectral information content. The approach is not limited to blood component analysis but has potential applications across diverse spectroscopic domains, providing new ideas and methods for improving the accuracy of quantitative spectroscopic analysis.