Hyperspectral remote sensing image classification method based on sparse graph regularization

Abstract

The invention relates to a hyperspectral remote sensing image classification method based on sparse graph regularization. The SGR method comprises the following steps: employing an effective sparse representation algorithm for a dictionary to obtain a score matrix, estimating a prediction function through optimizing a typical graph-based regularization problem, and finally, propagating labels of alarge amount of unknown data based on the score matrix and the prediction function to realize classification. Furthermore, a total variation sparse graph regularization method and a weighted joint sparse graph regularization method are introduced, total variation sparse representation classification TVSGR is expansion of the SGR in the aspect of models, spatial information modeled by using totalvariation is introduced in the sparse representation process, weighted sparse graph regularization classification WSGR is amplification of the SGR in the aspect of characteristics, and a Gaussian kernel-based adaptive weighting method is introduced into an SGR model; and the design method provided by the invention has very outstanding performance in the aspects of calculation efficiency, classification accuracy and noise robustness.

Description

technical field [0001] The invention relates to a hyperspectral remote sensing image classification method based on sparse graph regularization, belonging to the technical field of remote sensing image classification. Background technique [0002] Hyperspectral sensors can capture continuous spectral curves in a large number of electromagnetic bands. The so-called hyperspectral refers to its spectral resolution at the nanometer level, and the range can cover ultraviolet light, visible light, near-infrared, mid-infrared and thermal infrared. Hyperspectral remote sensing image (Hyperspectral Remote Sensing Image, HSI) is data with a "cube" structure. Different from traditional remote sensing images, hyperspectral remote sensing has the advantages of (1) small and numerous bands, (2) fine and continuous spectrum, (3) integration of spectrum and image, and (4) large amount of data. Therefore, hyperspectral remote sensing images combined with spectroscopy and imaging technology ...

AI Technical Summary

Problems solved by technology

Wang et al. improved on the basis of AGR and proposed two methods: efficient anchor graph regularization (Efficient AGR, EAGR) and hierarchical anchor graph regularization (Hierarchical AGR, HAGR), which further improved the classification accuracy. However, the AGR method has not been directly used in the research of hyperspectral classification

Images