Classification method and device for selecting high spectral wavelength based on single factor variance analysis

Abstract

The invention provides a classification method and device for selecting hyperspectral wavelengths based on single-factor variance analysis, and belongs to the technical field of image processing and spectroscopy. The classification method for selecting the hyperspectral wavelength based on the single factor variance analysis comprises the following steps: acquiring a hyperspectral image of a to-be-classified sample; carrying out image processing on the hyperspectral image of the sample to obtain the spectral information of the sample; correcting the spectral information of the sample by adopting standard normal variable transformation; calculating the characteristic wavelength in the spectral information of the sample by adopting single-factor variance analysis; extracting the spectral information of the sample corresponding to the characteristic wavelength from the corrected spectral information of the sample; dividing spectral information of samples corresponding to the characteristic wavelengths into a training set, a verification set and a test set; obtaining a sample classifier according to the spectral information training set and the verification set of the sample; and adopting a sample classifier to classify the spectral information test set of the samples. According to the method, the classification accuracy can be improved, and the classification modeling time is shortened.

Description

technical field [0001] The invention relates to the technical fields of image processing and spectroscopy, in particular to a classification method and device for selecting hyperspectral wavelengths based on single-factor analysis of variance. Background technique [0002] In recent years, near-infrared hyperspectral image technology has been applied in sample classification. Compared with single near-infrared spectroscopy technology, it contains richer spatial information and can realize non-destructive batch collection of spectral information of non-uniform experimental samples. However, the near-infrared hyperspectral image system has a high resolution, a small band gap between different wavelengths, and has a strong correlation, and the information redundancy is very strong. Therefore, efficient spectral wavelength selection methods are an important issue of research in this field. [0003] Aiming at the problem of high dimensionality of near-infrared data, some researc...

AI Technical Summary

Problems solved by technology

[0003] Aiming at the problem of high dimensionality of near-infrared data, some researchers use PCA to select the peaks and valleys in the principal component coefficients for wavelength selection, but PCA only retains the features that contribute the most to the variance in the data set during the mapping process, and does not use any Classification information inside the data

Some researchers have also proposed the continuous projection algorithm to be used in the quantitative analysis model of near-infrared spectroscopy to study the component content of the experimental sample, and achieved good results, but the method is still in general when transferred to the qualitative analysis model.

Some researchers also use modern genetic algorithms for wavelength selection. This method continuously evaluates the performance of the classifier and selects a subset of wavelengths until convergence. Although a relatively good local solution can be obtained, it takes too long

Images