Hyperspectral image classification method based on spatial and spectral features and sparse representation

Abstract

The invention aims at providing a hyperspectral image classification method based on spatial and spectral features and sparse representation. The method comprises the following steps that: hyperspectral high-dimension data is read, dimension conversion is carried out, and the hyperspectral high-dimension data is subjected to normalization processing, wherein the contained sample class number is L; the image spatial vein features and the spectral features are respectively extracted to obtain the spatial vein features T1 and the spectral features T2; a spatial and spectral feature set T of images is obtained through merging the spatial vein features and the spectral features, wherein T={T1, T2}; a part of samples is respectively selected from T for L classes to form a training set A, and a test set is set to a state that the set of all of the L classes in the T is M; the M and A are utilized for solving a sparse representation coefficient S<^> of the hyperspectral data for image reconstruction, and in addition, corresponding redundancy in each class is calculated; and the classes of the samples are determined according to the redundancy. The hyperspectral image classification method has the advantages that the information in hyperspectral images can be sufficiently utilized, and the hyperspectral images can be perfectly depicted through the spatial and spectral features; the classification precision can be improved; and the method can be applicable to different hyperspectral images, and the applicability is high.

Description

technical field [0001] The invention relates to an image processing method, specifically a hyperspectral image processing method. Background technique [0002] Hyperspectral images have very high spectral resolution, can obtain very narrow band intervals and a large number of spectral bands, so as to obtain image data containing rich information and continuous spectra, so they are widely used in agricultural production, mineral mapping, Target recognition and detection, disaster early warning and urban planning and other fields. Due to the large amount of data, high redundancy, high dimensionality, and strong correlation between bands are the main characteristics of hyperspectral data, which will bring great challenges to subsequent processing. [0003] Classification technology occupies a very important position in the process of hyperspectral image processing, which mainly divides the pixels with similar properties or characteristics in the data into the same class. Most...

AI Technical Summary

Problems solved by technology

Most of the existing classification methods consider the extraction and utilization of spectral information. The traditional hyperspectral data classification methods have the following problems: 1. The spatial information of the hyperspectral image itself is not fully utilized.

2. The classification accuracy is not high

Images