Nonlinear spectrum similarity measurement method

Abstract

The invention is a kind of nonlinear spectrum similarity measuring method, which belongs to information technology field. The invention uses kernel principal constituent analysis method to eliminate the strong correlation between spectrum wavebands: aiming at the strong correlation between spectrum wavebands, uses non-linear transformation to transform the original spectrum vectors into a new character space, in which the invention uses kernel function and principal analysis method to acquire the nonzero characteristic numbers and characteristic vectors of the covariance matrix, and arranges the characteristic numbers in descending order, selects the characteristic vectors correspondent to the front N characteristic numbers, forms a standard orthogonal basis, finally, the spectrum vectors in the characteristic space are projected onto the standard orthogonal basis; combines the spectrum reflection and absorption characters: multiplies the transferred reflecting rate vectors and the normalized absorption rate vector, forms a compound spectrum vector.

Description

technical field [0001] The invention relates to a spectral measurement method, in particular to a nonlinear spectral similarity measurement method based on the reflection and absorption characteristics of the complete waveform of the spectrum. It belongs to the field of information technology. Background technique [0002] According to literature search, spectral similarity measurement methods can be divided into two categories, one is based on complete waveform features, typical methods such as spectral angle mapping. Kruse, FA and others published "Spectral Image Processing System - Interactive Visualization and Analysis of Spectral Imager Data" (1993b, The Spectral Image Processing System (SIPS) - Interactive Visualization and Analysis of Imaging Spectrometer Data. Remote Sensing nvironment, 44:145-163), this paper proposes a spectral angle mapping method, which uses the angle between the spectral reflectance vectors to measure the similarity between the spectra. The ad...

AI Technical Summary

Problems solved by technology

This method is suitable for judging the similarity of specific absorption peaks in the spectrum. This method itself is not suitable for the similarity measurement between arbitrary spectra. However, this method reflects that different bands of the spectrum contribute differently to the spectral similarity measurement. Its The disadvantage is that it ignores the reflection characteristics of the spectrum and cannot be applied to the complete waveform of the spectrum

The reflection and absorption features of the spectrum are two types of features that cannot be separated from the spectrum. However, the current spectral similarity measurement method only targets one of the features.

In addition, the current spectral similarity methods all assume that the spectral bands are independent of each other. However, this assumption does not meet the actual situation. At present, the strong correlation between spectral bands is widely used in the dimensionality reduction and feature extraction of spectral data.

Images