A Hyperspectral Dimensionality Reduction Method Based on Optimal Graph Theory

Abstract

Provided is a hyperspectral dimension reducing method based on a map optimizing theory. The hyperspectral dimension reducing method based on the map optimizing theory solves the problems in the prior art that hyperspectral dimension reducing just depends on local neighbor information, non-neighbor difference is not considered, and subsequent classifying precision is not improved significantly. The hyperspectral dimension reducing method based on the map optimizing theory comprises the following steps of: 1) obtaining an optimized neighbor graph through nearest neighbor optimizing according to similarity measurement between original hyperspectral image pixels; 2) connecting any two non-neighbor pixels in the optimized neighbor graph with edges, obtaining a connecting graph, putting weights on the edges in the connecting graph, and forming a weight matrix; 3) ensuring a maximum distance between the non-neighbor pixels connected by edges after the dimension reducing when the optimized neighbor graph is mapped into a low dimension space; 4) solving character vectors and forming a conversion matrix; and 5) calculating a result of the original hyperspectral data after dimension reducing. According to the invention, the subsequent classifying precision of the hyperspectral images is improved.

Description

technical field [0001] The invention relates to the technical field of remote sensing, in particular to a dimensionality reduction method for removing redundant information from hyperspectral remote sensing data and enhancing differences between different types of ground targets. Background technique [0002] Hyperspectral images provide rich spectral information for distinguishing subtle differences between different ground targets. However, for hyperspectral image classification, not all band information is helpful for identifying ground targets, and a large number of spectral bands is a challenge for subsequent processing. Therefore, hyperspectral images often need dimension reduction processing before classification, which reduces the complexity of subsequent processing while retaining most of the important information. [0003] Due to the multidirectional scattering of hyperspectral images during the imaging process, the data inherently presents a nonlinear structure. ...

AI Technical Summary

Problems solved by technology

[0010] Aiming at the above problems, the present invention provides a hyperspectral dimensionality reduction method based on optimized map theory, so as to solve the problem that the dimensionality reduction of hyperspectral data in the prior art only depends on local neighbor information, without considering the difference of non-neighbors, which affects the subsequent classification accuracy. no significant improvement

Images