空间谱残差密集网络在压缩高光谱图像重建中的应用

134 浏览量更新于2024-08-27 1 收藏 903KB PDF 举报

"基于空间谱残差密集网络的压缩高光谱图像重建" 本文是一篇研究论文，探讨了利用空间谱残差密集网络进行压缩高光谱图像（HSI）重建的技术。高光谱图像拥有丰富的空间和光谱信息，但其大数据量特性使得存储和传输成为挑战，因此压缩技术显得尤为重要。在该研究中，作者提出了一种创新方法，旨在通过深度学习网络提高压缩后图像的重建质量。该方法主要包含两个核心网络：用于高光谱图像重建的残差密集网络（RDNHIR）和光谱差异重建网络（SDRN）。残差密集网络（RDNHIR）由多个残差密集块（RDBs）组成，其设计目的是通过级联所有RDBs的特征来提取图像的局部特征和全局层次特征。这种结构允许网络在深层学习过程中有效地捕获和传递信息，避免梯度消失或爆炸的问题。另一方面，光谱差异重建网络（SDRN）专注于利用高光谱图像中相邻波段之间的强相关性来更好地保留光谱特性。由于高光谱图像的光谱通道之间通常存在高度的相关性，SDRN可以更精确地恢复图像的光谱信息，从而改善重建效果。为了进一步提升性能，研究者在损失函数中引入了相邻光谱差异正则化。这一策略可以帮助优化网络训练，使其更加关注保持原始图像的光谱差异，从而提高重建图像的保真度和真实性。实验结果证明了所提方法的有效性，与现有技术相比，它在重建质量和计算效率方面都有显著优势。这表明，基于深度学习的空间谱残差密集网络在压缩高光谱图像重建领域具有广阔的应用前景，对于遥感、环境监测、军事侦察等领域的高光谱数据处理具有重要意义。

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

IEEE GEOSCIENCE AND REMOTE SENSING LETTERS 1

Compressive Hyperspectral Image Reconstruction

Based on Spatial–Spectral Residual

Dense Network

Wei Huang, Member, IEEE,YangXu , Member, IEEE, Xiaowei Hu, and Zhihui Wei

Abstract— A spatial–spectral residual dense network-based

compressiv e hyperspectral image (HSI) reconstruction method

is proposed in this letter. The proposed method contains two

networks: residual dense network for hyperspectral image recon-

struction (RDNHIR) and spectral d ifference reconstruction net-

work (SDRN). The RDNHIR network can extract the local

features and global hierarchical featur es by cascading features

of all residual dense blocks (RDBs). Then, SDRN takes full

advantage of the strong correlation between spectral adjacent

bands to better preserve the spectral feature of HSI. Finally,

the adjacent spectral difference regularization is introduced

into the loss function to further improve the performance. The

experimental results show that the proposed method has better

reconstruction quality than other state-of-the-art reconstruction

methods, especially in the spectral domain.

Index Terms— Hyperspectral image (HSI), residual dense

network, reconstruction, compressive sensing (CS)

I. INTRODUCTION

YPERSPECTRAL image (HSI) is a 3-D image with

a large amount of data composed of a series of band

images, which provides rich spatial and spectral information.

However, such a large amount of data may impose a heavy

burden on the storage and transmission of HSI. Compressive

sensing (CS) theory introduces a signal acquisition and recon-

struction framework that goes beyond the traditional Nyquist

sampling paradigm [1]. Therefore, CS theory is introduced to

reconstruct the HSI with a small amount of observation data.

HSI reconstruction aims to reconstruct the 3-D HSI from

its measurements, which is an underdetermined problem.

In order to solve this problem, image priors, such as total

variation (TV) prior [2] and sparsity prior [3], play a crucial

role. Later, Tan et al. [4] proposed a compressive HSI recon-

struction method via approximate message passing (AMP),

Manuscript received April 29, 2019; revised June 18, 2019; accepted

July 7, 2019. This work was supported in part by the National Natural

Science Foundation of China under Grant 61602423, Grant 61605175, and

Grant 61701238, in part by the Henan Province Science and Technology

Breakthrough Project under Grant 172102410088, in part by the Jiangsu

Provincial Natural Science Foundation of China under Grant BK20170858,

and in part by the Fundamental Research Funds for the Central Universities

under Grant 30919011234. (Corresponding author: Yang Xu.)

W. Huang is with the School of Computer and Communication Engineering,

Zhengzhou University of Light Industry, Zhengzhou 450000, China (e-mail:

hnhw235@163.com).

Y. Xu, X. Hu, and Z. Wei are with the School of Computer Science

and Engineering, Nanjing University of Science and Technology, Nan-

jing 210094, China (e-mail: xuyangth90@njust.edu.cn; 574801016@qq.com;

gswei@njust.edu.cn).

Color versions of one or more of the ﬁgures in this letter are available

online at http://ieeexplore.ieee.org.

Digital Object Identiﬁer 10.1109/LGRS.2019.2930645

which used an adaptive Winer ﬁlter as the image denoiser in

each iteration. By doing so, the AMP method can produce bet-

ter results and has the advantage of no parameter adjustment.

Metzler et al. [5] developed an extension of AMP framework

named denoising-based AMP (D-AMP). It can integrate a

wide class of denoisers within its iterations. Although the

above methods can reconstruct the HSI with better spatial

structure, they do not take the spectral information of HSI

into consideration. Martin et al. [6] proposed a new technique

named hyperspectral coded aperture (HYCA). It is based

on the assumption that the spectral vectors live on a low-

dimensional subspace and the spectral bands present high

correlation in both the spatial and the spectral domains and

can obtain high compression ratios with very small errors.

Wang et al. [7] presented a CS reconstruction algorithm with

spectral unmixing for HSI, which not only considered the

spatial information but also took full advantage of spec-

tral mixing property. However, these methods need to be

solved iteratively, and the computational complexity is very

high.

Recently, deep learning-based methods have great repre-

sentational power for a complex feature and have achieved

success in the ﬁeld of remote sensing image processing, such

as multispectral processing [8] and HSI processing [9]–[13].

Xiong et al. [14] proposed a uniﬁed deep learning frame-

work to reconstruct HSI from spectrally undersampled projec-

tions. Speciﬁcally, a coded aperture snapshot spectral imaging

(CASSI) measurement is ﬁrst “upsampled” to the desired

spectral resolution using a simple CS reconstruction algo-

rithm and then enhances the image details by CNN model.

Choi et al. [15] employed the convolutional autoencoder to

reconstruct its own input. The spectral feature is learnt through

the encoder and decoder networks. The regularization with

the sparsity of the gradients in the spatial is also introduced.

Mousavi and Baraniuk [16] developed a new signal recovery

framework named DeepInverse that learns the inverse trans-

formation from measurement vectors to signals using a deep

convolutional network. Hu et al. [17] presented patch-based

residual networks for HSI reconstruction, and it contains two

residual convolutional neural networks: one is reconstruction

network for CS reconstruction and the other is deblocking

network for removing the blocky effect. This method can efﬁ-

ciently reconstruct all bands of HSI jointly to better preserve

spectral correlation. These methods based on a deep network

can be used for HSI reconstruction, but they do not make full

use of the hierarchical features.

To solve this problem, we propose a compressive HSI

reconstruction method based on a spatial–spectral residual

See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

Authorized licensed use limited to: Zhengzhou Institute of Light Industry. Downloaded on April 10,2020 at 00:49:25 UTC from IEEE Xplore. Restrictions apply.

下载后可阅读完整内容，剩余4页未读，立即下载

weixin_38624557

粉丝: 8
资源: 912

空间谱残差密集网络在压缩高光谱图像重建中的应用

利用残差密集网络的高光谱图像分类

基于残差密集网络层次信息的图像标题生成.pdf

基于补丁的残差网络用于压缩感知的高光谱图像重建

基于多尺度残差网络的小样本高光谱图像分类

残差密集网络提升高光谱图像分类精度

非迭代补丁残差网络：高光谱图像压缩感知重建新方法

残差网络用作高光谱图像分类

基于对抗式残差密集深度神经网络的CT稀疏重建.docx

基于三维空洞卷积残差神经网络的高光谱影像分类方法

PZRes-Net:论文“通过深度渐进零中心残差学习实现的高光谱图像超分辨率”的源代码

最新资源