Spatial-Spectral Deep Residual Network for Hyperspectral Image Super-Resolution

Recently, single hyperspectral image super-resolution (SR) methods based on deep learning have been extensively studied. However, there has been limited technical development focusing on single hyperspectral image super-resolution due to the high-dimensional and complex spectral patterns in hyperspe...

Full description

Saved in:
Bibliographic Details
Published in:SN computer science 2023-07, Vol.4 (4), p.424, Article 424
Main Authors: Zheng, WeiFa, Xie, ZiXin
Format: Article
Language:English
Subjects:
Computer Imaging
Computer Science
Computer Systems Organization and Communication Networks
Data Structures and Information Theory
Deep learning
Dictionaries
High resolution
Hyperspectral imaging
Image resolution
Information Systems and Communication Service
Methods
Neural networks
Original Research
Pattern Recognition and Graphics
Software Engineering/Programming and Operating Systems
Spatial data
Vision
Citations: Items that this one cites
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Recently, single hyperspectral image super-resolution (SR) methods based on deep learning have been extensively studied. However, there has been limited technical development focusing on single hyperspectral image super-resolution due to the high-dimensional and complex spectral patterns in hyperspectral image. Besides, most existing methods can not effectively explore spatial information and spectral information of hyperspectral image, obtaining relatively low performance. To address these issues, in this paper, we propose a novel spatial-spectral deep residual network (SSDRN) for hyperspectral image super-resolution. To fully exploit the spectral information and spatial correlation characteristics in hyperspectral data, we use residual group (RG) to extract features effectively, which consists of the proposed spatial-spectral residual attention block (SSRAB). Furthermore, we present a upsampling reconstruction module (UP) to utilize the deep features of previously obtained low-resolution hyperspectral image. Then we could learn the nonlinear mapping relationship between low resolution and high resolution to obtain the desired high-resolution hyperspectral image. Experimental results on two benchmark datasets demonstrate that the proposed approach achieves superior performance in comparison to existing state-of-the-art methods.
ISSN:2661-8907
2662-995X
2661-8907
DOI:10.1007/s42979-023-01868-0