Multi-model fused spectral wavelength selection method

Abstract

The invention relates to a spectral wavelength selection method, which comprises the following steps of: dividing a spectrum into a plurality of wave band intervals; carrying out partial least square regression analysis on each wave band interval; calculating an interactive verification mean square error of a variable corresponding to each wave band; finding out wavebands with small interactive verification mean square error, and combining the spectral absorption matrixes corresponding to the waveband intervals to obtain a new spectral absorption matrix; for the new spectral absorption matrix, using a Monte Carlo sampling method to carry out multiple times of sampling, removing wavelength points with relatively small regression coefficients in each time of sampling, then establishing a partial least square regression analysis model, and selecting a wavelength variable set with a minimum interactive verification mean square error as a candidate optimal wavelength variable set; and carrying out multiple rounds of sampling to select a repeated and stable wavelength variable as an optimal wavelength variable. According to the method, the stable optimal spectral wavelength variables are selected through multiple rounds of sampling and screening, and compared with a single spectral wavelength selection model, the selected spectral wavelength variables are fewer.

Description

technical field [0001] The invention belongs to the field of spectral analysis of material components, and in particular relates to a multi-model fusion spectral wavelength selection method. Background technique [0002] Spectral analysis is an emerging material analysis technology, which has been widely used in agriculture, medical treatment, environmental monitoring and other fields due to its advantages of fast, non-destructive, and no secondary pollution. The types of spectral data used in spectral analysis mainly include Raman spectroscopy, infrared spectroscopy, fluorescence spectroscopy, and ultraviolet spectroscopy. In the detection of mixtures, these kinds of spectra will obtain a large number of wavelength points and the spectral absorption peaks of different substances will overlap, and the wavelength variables will have serious collinearity. These characteristics will affect the accuracy of subsequent qualitative and quantitative analysis. Therefore, it is of g...

AI Technical Summary

Problems solved by technology

[0007] The above methods all use a model for wavelength selection. After wavelength selection in the full spectrum range, there are still many wavelength variables, and more sample training sets are required.

Images