Method for selecting characteristic wavelength of near-infrared spectrum based on simulated annealing algorithm

Abstract

The invention discloses a method for selecting characteristic wavelength of a near-infrared spectrum based on a simulated annealing algorithm, which comprises the following steps: selecting the minimum number of wave number points and repeatedly selecting k wave number points randomly; establishing initial PLS models of a correction set and a prediction set, and calculating a cross-validation root mean square error value of a corresponding initial PLS model; selecting optimal k wave number points highly associated with a chemical value of the correction set from the full spectrum by using the simulated annealing algorithm, such that the window width is dynamically increased progressively to ensure the obtaining of global optimal solution of the simulated annealing algorithm; when finishing, comparing the modeling effects corresponding to the wave number point sets under the condition of different window widths to obtain the optimal window width and the optimal wave number point combination, and establishing the final PLS model. The method eliminates the subjective factor influence resulting from manual designation of window width through dynamic increasing of the window width; and by judging the importance of the newly selected wave number point according to the Metropolis rule, the method ensures the rapid convergence of algorithm, avoids the problem that the simulated annealing algorithm only realizes local optimal solution, and has higher reliability and precision.

Description

technical field [0001] The invention relates to a method for selecting characteristic wavelengths of near-infrared spectra, in particular to a method for selecting characteristic wavelengths of near-infrared spectra based on a simulated annealing algorithm for detection of agricultural products or food. Background technique [0002] Near-infrared spectroscopy is a high-speed, accurate and green spectral analysis technology, which is widely used in agricultural product quality inspection and food analysis. The basic principle of near-infrared spectroscopy technology is that the near-infrared spectrum contains the frequency multiplication and combination frequency information of the fundamental frequency vibration of a single chemical bond in the molecule, mainly the frequency multiplication and combination frequency vibration of the hydrogen-containing group X-H (H is C, N, O). overlay. Correlating the spectrum with the sample quality parameters, determining the quantitative...

AI Technical Summary

Problems solved by technology

Near-infrared spectroscopy technology has many advantages, but its disadvantages are: near-infrared spectroscopy technology is an indirect analysis technique, and the spectrum contains a lot of information, but the information intensity is low and the spectral peaks overlap. Selecting the characteristic wavelengths that are highly correlated with the same sample is the key to obtaining the optimal near-infrared spectrum model. In order to overcome the above shortcomings, it is necessary to select the characteristic wavelengths of the near-infrared spectrum

This method can be used in practical applications, but the disadvantages of this method are: the division of intervals is based on personal experience and lacks scientific basis; it is assumed that the characteristic wavelengths are concentrated in a certain sub-interval, and the sub-interval All wavenumber points are characteristic wavenumber points, that is, there is no trade-off between wavenumber points in the subinterval, and it is difficult to satisfy the above assumptions in the actual modeling process

Using permutation and combination to solve, although it can overcome the deficiency of interval partial least squares method to select wavelength sub-intervals, but this solution method has a huge amount of calculation, and the amount of calculation increases exponentially with the increase of the wavenumber points of the full spectrum, which cannot be applied to actual modeling. process

Images