Hyperspectral Image Super-Resolution Reconstruction Method Based on Spectral Space Combination and Gradient Domain Loss

Abstract

The invention relates to a hyperspectral image super-resolution reconstruction method based on spectrum-space combination and gradient domain loss. The method includes steps: S1, obtaining a hyperspectral image; S2, dividing the hyperspectral image into a training set and a test set; S3, inputting the training set to a neural network with spectrum-space combination, and performing training by employing a combined loss of a spatial domain and a gradient domain; and S4, enabling the test set to pass through the neural network, and obtaining a final reconstruction result. According to the adoptedmethod, compared with the prior art, light weight is achieved in a network structure, the reconstruction quality is higher, and the anti-noise performance is great.

Description

Technical field[0001]The invention belongs to the field of image super-resolution reconstruction, and relates to a hyperspectral image super-resolution reconstruction method based on spectral space combined network and gradient domain loss.Background technique[0002]The low spatial resolution of hyperspectral images makes it easy to produce mixed end members, resulting in spectral distortion and destroying the spatial and spectral consistency of end members. In terms of input information, super-resolution reconstruction of hyperspectral images can be roughly divided into two types: reconstruction methods based on image fusion (with auxiliary images) and reconstruction methods without auxiliary images. The reconstruction method based on image fusion uses RGB images or panchromatic (PAN) images or multispectral (MS) images as an aid, and uses spatial and spectral information to jointly constrain, thereby unmixing endmembers and reducing spectral distortion. Using the RGB-assisted recon...

AI Technical Summary

Problems solved by technology

[0005] 1) The time complexity and reconstruction quality of the method cannot be balanced;

[0006] 2) Existing methods cannot capture the spectral information of hyperspectral images to combine spectrum and space, and the reconstruction quality is poor

Images