Unsupervised Hyperspectral Image Blind Fusion Method and System Based on Space-Spectrum Joint Residual Correction Network

Abstract

The invention discloses an unsupervised hyperspectral image blind fusion method and system based on a spatial-spectral joint residual correction network. The method includes: establishing a degraded network structure of a hyperspectral image to simulate the process of spatial and spectral downsampling; establishing a spatial and the spectral residual fusion network model, using the difference between the low-resolution results obtained by the degradation model and the training data as the input of the fusion network, that is, the residuals in the spatial and spectral dimensions are fused to obtain a residual map corresponding to the input data ; Correct the initialization data, and then send the corrected result to the degradation network and the space-spectrum joint correction network for multiple iterations to improve the accuracy of the fusion result. The present invention uses a space-spectrum joint correction network suitable for blind fusion of unsupervised hyperspectral images, and the space-spectrum joint correction network can obtain an error map between an input hyperspectral image and a real value.

Description

technical field [0001] The invention belongs to the technical field of remote sensing image processing, and in particular relates to an unsupervised hyperspectral image blind fusion method and system based on a space-spectrum joint residual correction network. Background technique [0002] Hyperspectral image fusion is an important application direction in the field of hyperspectral image remote sensing. Hyperspectral image fusion is to use the rich spectral information in low spatial resolution hyperspectral images and the rich spatial information in high spatial resolution multispectral images to synthesize high spatial resolution hyperspectral image data, which can be used for subsequent more complex images. Processing tasks provide high-quality training sets. Due to the limitations of existing sensor hardware, it is difficult to directly acquire images with both high spatial resolution and high spectral resolution. Therefore, the collected data can be post-processed thr...

AI Technical Summary

Problems solved by technology

[0004] However, the hyperspectral image fusion based on deep learning still has the following problems: (1) The two parameters of the spatial downsampling operator and the spectral response matrix depend on the imaging device, and many super-resolution methods assume these two operators by simulating the information of the imaging device It is known that the greater the error between the assumed operator and the actual parameter information, the more the accuracy of the fusion result will deviate from the real hyperspectral image; (2) The image obtained by the imaging device often reduces the resolution of one dimension to improve the other. One-dimensional resolution, that is, it is difficult to directly obtain high-spatial-resolution hyperspectral images, so building a supervised training model does not meet actual needs; (3) Unknown spatial degradation operators and spectral response matrices are unknown. Supervised fusion is a highly ill-posed problem, because unsupervised training requires that the loss function does not use real high-spatial-resolution hyperspectral images, and can only evaluate image quality indirectly, such as using low-resolution images after output degradation and known How to use limited low-resolution data to train an unsupervised blind fusion network and design an effective loss function is a difficulty

Images