[go: up one dir, main page]
 
 
Search for Articles:
Title / Keyword
Author / Affiliation / Email
Journal
Article Type
 
 
Advanced Search
 
Section
Special Issue
Volume
Issue
Number
Page
 

Saved Queries

Sign in to use this feature.

Search Filter Reset All

Years

Between: -

Subjects

remove_circle_outline
remove_circle_outline
Add Subjects Reset
Select Subjects

Journals

remove_circle_outline
Add Journals Reset
Select Journals

Article Types

remove_circle_outline
Add Article Types Reset
Select Article Types

Countries / Regions

remove_circle_outline
Add Countries / Regions Reset
Select Countries / Regions
Update Search
share announcement

Search Results (1)

Search Parameters:
Keywords = TRQ3DNet

Order results
Result details
Results per page
Show export options expand_more Show export options expand_less
Select all
Export citation of selected articles as:
20 pages, 22357 KiB  
Article
TRQ3DNet: A 3D Quasi-Recurrent and Transformer Based Network for Hyperspectral Image Denoising
by Li Pang, Weizhen Gu and Xiangyong Cao
Remote Sens. 2022, 14(18), 4598; https://doi.org/10.3390/rs14184598 - 14 Sep 2022
Cited by 31 | Viewed by 2831
Abstract
We propose a new deep neural network termed TRQ3DNet which combines convolutional neural network (CNN) and transformer for hyperspectral image (HSI) denoising. The network consists of two branches. One is built by 3D quasi-recurrent blocks, including convolution and quasi-recurrent pooling operation. Specifically, the [...] Read more.
We propose a new deep neural network termed TRQ3DNet which combines convolutional neural network (CNN) and transformer for hyperspectral image (HSI) denoising. The network consists of two branches. One is built by 3D quasi-recurrent blocks, including convolution and quasi-recurrent pooling operation. Specifically, the 3D convolution can extract the spatial correlation within a band, and spectral correlation between different bands, while the quasi-recurrent pooling operation is able to exploit global correlation along the spectrum. The other branch is composed of a series of Uformer blocks. The Uformer block uses window-based multi-head self-attention (W-MSA) mechanism and the locally enhanced feed-forward network (LeFF) to exploit the global and local spatial features. To fuse the features extracted by the two branches, we develop a bidirectional integration bridge (BI bridge) for better preserving the image feature information. Experimental results on synthetic and real HSI data show the superiority of our proposed network. For example, in the case of Gaussian noise with sigma 70, the PSNR value of our method significantly increases about 0.8 compared with other state-of-the-art methods. Full article
(This article belongs to the Special Issue Machine Vision and Advanced Image Processing in Remote Sensing)
►▼ Show Figures

Figure 1

Figure 1
<p>The overall architecture of the TRQ3DNet.</p> Full article ">Figure 2
<p>The structure of the 3D quasi-recurrent block.</p> Full article ">Figure 3
<p>(<b>a</b>) An overview of the Uformer block. (<b>b</b>) Calculation method of window-based multi-head self-attention (W-MSA). (<b>c</b>) The structure of the locally-enhanced feed-forward network (LeFF).</p> Full article ">Figure 4
<p>Gaussian (<math display="inline"><semantics> <mrow> <mi>σ</mi> <mo>=</mo> <mn>50</mn> </mrow> </semantics></math>) denoising outputs at the 20th band of the image on the ICVL dataset.</p> Full article ">Figure 5
<p>Visual comparison of the denoised image for all the five cases of the ICVL dataset.</p> Full article ">Figure 6
<p>PSNR values of each band for the synthetic experiments.</p> Full article ">Figure 7
<p>The denoised results on band 2 of the Indian Pines dataset for all the competing methods.</p> Full article ">Figure 8
<p>The denoised results on band 106 of the Urban dataset for all the competing methods.</p> Full article ">Figure 9
<p>PSNR comparison of different training strategies.</p> Full article ">Figure 10
<p>Comparison of feature maps of QRU3D, Uformer, and TRQ3D.</p> Full article ">Figure 11
<p>Visualization of different network structures. Case (<b>a</b>) and case (<b>b</b>) place the BI bridge in the input and the output of the unit, respectively, and case (<b>c</b>) places one way of the bridge in the input and the other way in the output. In addition, there is only one direction of information exchange in case (<b>d</b>) and case (<b>e</b>) replaces 3D convolution with 2D convolution in each TRQ3D unit.</p> Full article ">
Show export options expand_more Show export options expand_less
Select all
Export citation of selected articles as:
 
Displaying article 1-50 on page 1 of 1.
Back to TopTop