Multi-hyperspectral image fusion method guided by low-rank prior and spatial spectrum information

Abstract

The invention discloses a multi-hyperspectral image fusion method guided by low-rank prior and spatial spectrum information, and provides a brand new multi-layer multi-branch fusion network SSLRNet combining spatial spectrum guidance and low-rank prior, the network firstly constructs a multi-layer multi-branch fusion sub-network (MLMB), and aims to extract features from a plurality of branches, and then the multi-layer multi-branch fusion sub-network SSLRNet constructs a multi-layer multi-branch fusion sub-network SSLRNet; and multi-layer feature fusion is carried out to reconstruct a preliminary fusion image. And then, constructing a fusion image spatial spectrum correction sub-network based on spatial spectrum guidance, and performing spatial spectrum guidance on a preliminary fusion image generated by the MLMB through a multispectral image waveband superposition summation image and a hyperspectral image waveband average value image, thereby reducing spatial spectrum distortion. And finally, constructing a fusion image low-rank prior constraint sub-network based on a low-rank neural network, combining the sub-network with a deep learning network, and performing low-rank decomposition by utilizing the characteristics of the network, so that a fusion result better meets a real application requirement. According to the invention, the fusion precision of the network is improved, and real application requirements are better met.

Description

technical field [0001] The invention relates to the field of hyperspectral image and multispectral image fusion, in particular to a technical method for reducing spatial distortion and spectral distortion generated in the fusion process through spatial spectrum guidance, and embedding low-rank priors in a neural network to enable The fusion image is a technical method that is more in line with the needs of practical applications. It completes the mapping of low-resolution images to high-resolution images in a data-driven manner, and realizes the effective fusion of hyperspectral images and multispectral images. Background technique [0002] Hyperspectral images usually have dozens or hundreds of narrow spectral range bands, which means they have high spectral resolution. Using this feature, very delicate spectral curves can be simulated, and different types of ground objects and features can be identified. , so it is widely used in a series of tasks such as image classificat...

AI Technical Summary

Problems solved by technology

[0008] In the fusion method based on deep learning, hyperspectral images and multispectral images are usually superimposed as network input, which is difficult to preserve well, extracting the respective characteristics of multispectral and hyperspectral images, which increases the difficulty of fusion

At the same time, it lacks effective space-spectrum guidance and has serious space-spectrum distortion

Furthermore, the entire fusion process lacks physical interpretation, and it is easy to ignore the inherent prior characteristics of hyperspectral images, making the fusion images not necessarily meet the needs of real applications.

[0009] It can be seen that an ideal hyperspectral image and multispectral image fusion method has not yet appeared.

Images