.png){kind=icon}
.png){kind=icon}
Return
IATR-fuse: adaptive enhanced infrared and visible image fusion network based on improved DenseNet and transformer
.png){kind=icon}
PRE
AI
D
Z
DOI：10.1117/1.JEI.35.2.023040.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Infrared and visible image fusion aims to integrate thermal signatures from infrared images with high-frequency textural details from visible images, particularly under challenging illumination conditions. Existing deep-learning-based approaches predominantly rely on convolutional operations, constraining their ability to capture long-range contextual dependencies. To overcome this limitation, we propose IATR-Fuse, an end-to-end adaptive enhanced fusion network combining an Improved DenseNet Module (IMDesM) and transformer module (TRM). The IMDesM employs parallel dense connections for effective feature extraction, whereas TRM captures global spatial interdependencies via self-attention mechanisms. We further design an attention residual feature enhancement module (ARFEM) to refine the feature outputs by emphasizing salient information. A hybrid loss function, which integrates pixel fidelity, structural similarity, and gradient preservation with adaptive weight optimization, facilitates unsupervised learning. Extensive experiments on public datasets demonstrate that IATR-Fuse consistently achieves superior fusion performance, surpassing state-of-the-art methods in both qualitative and quantitative assessments while exhibiting strong generalization across diverse scenarios.
Keywords:
image fusion
transformer
DenseNet
adaptive
infrared
Journal
J
IF:
1
Papers:
109
Citations:
2.7K
Organization
Cited Papers