A hyperspectral classification method based on attention constraint non-negative matrix factorization

Abstract

The invention relates to a hyperspectral classification method based on attention constraint non-negative matrix factorization. The method comprises the following steps: S1, inputting original hyperspectral image data; S2, normalizing the hyperspectral image matrix by using the highlight to obtain a to-be-processed data set X; S3, decomposing X by adopting NMF to obtain an end member matrix U anda bottom abundance matrix H; S4, normalizing the end member matrix U and the bottom abundance matrix H; S5, initializing the attention parameterization matrix W according to the bottom abundance matrix H; S6, normalizing the attention parameterization matrix W; S7, the hyperspectral image, the end element matrix and the bottom abundance matrix are subjected to attention parameterization matrix, and an attention non-negative matrix is adopted to be decomposed, updated and iterated to be converged to obtain an end element matrix and a corresponding abundance matrix; According to the invention, the end member position information in the abundance matrix obtained when the non-negative matrix decomposition technology decomposes the hyperspectral image is not easy to lose, so that the classification precision of the hyperspectral image is improved.

Description

technical field [0001] The present invention relates to the field of hyperspectral classification, and more specifically, to a hyperspectral classification method based on attention-constrained non-negative matrix factorization. Background technique [0002] Hyperspectral remote sensing technology is a remote sensing information acquisition technology developed on the basis of imaging spectroscopy; it can obtain hundreds of continuous high-resolution images on the spectrum, and each pixel in the image corresponds to a spectral curve. The dimension of the spectral information contained in it is equal to the number of imaging frames; since the bands of the hyperspectral image are dense and overlap between them, a continuous radiation curve can be used to represent the characteristics of each pixel in the image data, correspondingly, The spectral curve of a group of pixels can be used to represent the distribution of ground features; because of its high spectral resolution and ...

AI Technical Summary

Problems solved by technology

However, the ground reflectance spectral signal obtained by hyperspectral remote sensing is recorded in units of pixels, which is a synthesis of the spectral signals of surface materials corresponding to the pixel; if the pixel contains only one type of ground object, such as minerals, Water bodies, vegetation, etc., are called end members; if the pixel contains more than one type of surface features, it is called a mixed pixel; there will be a large number of "same objects with different spectra" and "different objects with the same spectrum" in hyperspectral images. Spectrum” phenomenon, purely using spectral information for classification is very likely to cause misclassification of some ground object types; at the same time, with the improvement of spectral resolution of hyperspectral remote sensing data, its data dimension and data volume also increase significantly , for example, AVIRIS has 244 bands, which significantly increases the pressure on the computer when processing data, and brings difficulties to the identification and classification of hyperspectral; so the existing non-negative matrix factorization technology obtains the abundance matrix when decomposing hyperspectral images The position information of the endmember is easy to lose, resulting in low classification accuracy of hyperspectral images

Images