Hyperspectral image classification algorithm based on dual denoising in combination with multi-scale superpixel dimension reduction

Abstract

According to the method, a dual denoising algorithm based on transform domain filtering denoising and spatial domain filtering denoising is combined and applied to dimension reduction classification of the multi-scale hyperspectral image containing the superpixels, so compared with a hyperspectral image classification algorithm based on transform domain filtering denoising, the method employs hierarchical domain transformation recursive filtering for eliminating a small-scale texture structure, and meanwhile, an edge protection effect is achieved; compared with a hyperspectral image classification algorithm based on spatial domain filtering denoising, the method has the advantages that the noise part is separated from the original hyperspectral image by using the non-low rank attribute of the noise part, so mixed noise in the original image can be removed, the image quality is enhanced, and the subsequent classification precision of the hyperspectral image is improved.

Description

technical field [0001] The invention relates to the technical field of image processing, in particular to a hyperspectral image classification algorithm based on double denoising combined with multi-scale dimensionality reduction. Background technique [0002] Hyperspectral remote sensing refers to the science and technology of remote sensing data acquisition, processing, analysis and application with high spectral resolution. Usually, two types of sensors covering a certain spectral range, imaging spectrometer and non-imaging spectrometer, are used to obtain data. Hyperspectral image classification is an important part of remote sensing image processing and application, and its ultimate goal is to assign a unique identifier to each pixel in remote sensing images. However, hyperspectral images are often disturbed by various types of noise during the acquisition process, and the high-dimensional characteristics of hyperspectral images, high correlation between bands, spectral...

AI Technical Summary

Problems solved by technology

However, the classification effect of the algorithm for mixed noise and different intensity noise images is not ideal, and it cannot meet the requirements of noise robustness; the transform domain filtering algorithm based on low-rank sparse representation uses the low-rank sparse characteristics of hyperspectral images to remove mixed noise and restore the original image, to a certain extent, improves the classification accuracy of hyperspectral images, but cannot eliminate the small-scale structure in the image, making the classification effect unsatisfactory

Images