A Quadratic Feature Selection Method for Fluorescence Spectroscopy Based on Hybrid Particle Swarm and Continuous Projection

Abstract

The invention discloses a fluorescence spectrum secondary feature selection method based on mixed particle swarm and continuous projection, which is characterized in that: first, the mixed particle swarm algorithm is used to perform the first feature selection on the fluorescence spectrum, and the original high-dimensional fluorescence spectrum is The first dimensionality reduction is achieved by selecting fluorescence spectra with significant features in several dimensions; and then continuing to apply the continuous projection algorithm in the characteristic spectrum to achieve the second feature selection, so as to perform the second dimensionality reduction on the fluorescence spectrum. This method makes full use of the advantages of particle swarm algorithm, genetic algorithm and continuous projection algorithm, and can effectively reduce the dimensionality of fluorescence spectra.

Description

technical field [0001] The invention belongs to the field of spectral information processing, and relates to a feature dimension reduction processing method of fluorescence spectrum, in particular to a secondary feature selection method of fluorescence spectrum based on mixed particle swarm and continuous projection. Background technique [0002] The fluorescence spectrum of a substance has the characteristics of high dimensionality and large amount of information. Modeling and quantitative analysis based on the original fluorescence spectrum usually has high accuracy. However, due to the large amount of redundant information in the original spectrum, the model training is complicated and the calculation low efficiency. It is particularly necessary to reduce the dimensionality of the original fluorescence spectrum by using the method of feature selection. Modeling and analysis based on the feature spectrum after dimensionality reduction will greatly speed up the training spe...

AI Technical Summary

Problems solved by technology

[0003] There are still some shortcomings in the existing feature selection methods. For example, when using the particle swarm optimization algorithm, it uses the currently searched optimal position as shared information, and it is easy to fall into a local optimum, resulting in the phenomenon of "premature convergence". Usually, it needs to be combined with Other algorithms are improved to avoid premature convergence to local optimal solutions

After combining with other algorithms, the feature spectrum after dimensionality reduction usually still has a high dimension, the corresponding modeling variables still contain certain redundant information, and the model complexity is still high, which is not conducive to the improvement of the model training speed

Images