Band Selection Method for Hyperspectral Image Based on Discriminant Information and Manifold Information

Abstract

The present invention proposes a hyperspectral image band selection method based on discriminant information and manifold information, which mainly solves the problems that existing band selection methods fail to effectively utilize data manifold information, lose the original physical meaning of data and have error accumulation. The implementation steps are: 1) convert the hyperspectral image into a two-dimensional matrix Y; 2) normalize Y to obtain a two-dimensional matrix X; 3) construct a kernel function K and a linear discriminant expression according to X; 4) According to the result of 3), calculate the regular matrix G of the graph; 5) combine G to iteratively calculate the coefficient representation matrix W; 6) judge whether the number of iterations k is greater than or equal to the maximum number of iterations, if so, output the final W, and execute 7), otherwise, set k =k+1, return 5); 7) Calculate all band scores according to W to form a new matrix X * to complete the band selection. The invention reduces cumulative errors, and the selected wave bands are more representative, which can be used for preprocessing of hyperspectral images.

Description

technical field [0001] The invention belongs to the field of image processing, and specifically relates to a method for selecting bands of hyperspectral images, which can be used for preprocessing hyperspectral images involved in fields such as agriculture, geology, atmosphere and hydrology. Background technique [0002] Benefiting from the rapid development of hyperspectral imaging technology and the continuous improvement of imaging quality, hyperspectral images have been more and more widely used in fields such as agriculture, geology, atmosphere, and hydrology, and the information contained in images has become more and more abundant. On the other hand, rich image information also brings problems such as large data volume and redundant bands, which restrict the storage, transmission and processing of hyperspectral images. At present, many hyperspectral band selection methods have been proposed at home and abroad to solve the above problems. [0003] In their paper "Band...

AI Technical Summary

Problems solved by technology

However, the shortcoming of this method is that it fails to utilize the intrinsic manifold information of the data, which leads to the inaccuracy of the low-rank representation coefficients learned by it, which affects the judgment of the representativeness of the bands, that is, the importance of different bands may be wrongly judged size, the quality of the new image composed of the finally selected bands cannot be guaranteed

Images