Determine the significant digit of spectral data and reduce its redundant digits to eliminate the chance correlation problem based on the “salami slicing” method

Abstract

In recent years, complex solution composition analysis based on spectroscopy has been a research hotspot for researchers and has broad application prospects. Improving the ability of complex solutions component analysis based on spectroscopy, eliminating spectral data redundancy and the chance correlation problem it brings has become an urgent issue. In order to solve these problems, this paper takes the dynamic spectrum (DS) as the research object, and uses the “salami slicing” method to process the DS data. Firstly, the significant digit of the decimal DS data is processed, and then the partial least-squares (PLS) method is applied to model and analyze the processed DS data. The turning point of the modeling accuracy's change is found, and the significant digit number of DS data is determined roughly. On this basis, the weight of the binary number is used skillfully to process the significant digit of DS. The processed DS data is modeled, and the significant digit number of the DS data is accurately analyzed to establish the efficacy of the proposed method. This method does not only improve the signal-to-noise ratio (SNR) of DS data, but also avoids the chance correlation problem due to the data redundancy. This method also provides a good means for SNR estimation of other spectral data and avoiding the chance correlation problem in modeling, and has a high application value.