一个一般的图像融合方法利用梯度转移学习和融合规则展开
IEEE transactions on pattern analysis and machine intelligence
|January 19, 2026
概括
这项研究引入了一种全新的深度学习框架,用于一般图像融合,增强模型培训和网络设计. 该方法有效地利用跨任务的互补信息,为各种应用产生卓越的融合结果.
科学领域:
- 计算机视觉 计算机视觉
- 人工智能的人工智能
- 图像处理 图像处理
背景情况:
- 现有的深度学习图像融合方法在模型培训和网络设计方面缺乏效率.
- 目前的方法未能有效地利用各种融合任务中的互补信息.
- 基于启发式的网络设计限制了一般图像融合模型的多功能性.
研究的目的:
- 为一般图像融合提出一个全面的深度学习框架.
- 为解决单模型多任务融合的模型培训和网络设计方面的局限性.
- 为实际应用开发一个多功能和高效的图像融合网络.
主要方法:
- 开发了一个顺序梯度转移框架,以利用跨任务的互补信息.
- 拟议的融合规则展开,集成到网络设计的深平衡模型中.
- 利用梯度转移学习来加强训练期间的信息提取.
主要成果:
- 拟议的方法在多焦点,多曝光和红外/可见任务中实现了卓越的图像融合结果.
- 生成的图像显示了更丰富的结构信息和竞争性的客观指标.
- 在看不见的医疗图像融合任务中表现出显著的性能改善.
结论:
- 这种新的框架为一般的图像融合提供了一个高效和多功能解决方案.
- 梯度转移学习和融合规则的展开使有效的多任务学习成为可能.
- 该方法显示了强大的泛化能力,用于各种图像融合应用.
相关概念视频
Nuclear Fusion
33.7K
The process of converting very light nuclei into heavier nuclei is also accompanied by the conversion of mass into large amounts of energy, a process called fusion. The principal source of energy in the sun is a net fusion reaction in which four hydrogen nuclei fuse and ultimately produce one helium nucleus and two positrons.
A helium nucleus has a mass that is 0.7% less than that of four hydrogen nuclei; this lost mass is converted into energy during the fusion. This reaction produces about...
A helium nucleus has a mass that is 0.7% less than that of four hydrogen nuclei; this lost mass is converted into energy during the fusion. This reaction produces about...
33.7K
Tagging and Fusion Proteins
8.3K
Proteins are involved in several cellular processes and biochemical reactions. Analyzing a specific protein of interest requires it to be isolated from the other proteins in the cell. This is achieved by overexpressing the specific gene in a suitable host to produce large quantities of the target protein. A tag or label is recombined with the gene to produce a fusion protein containing the target protein and the tag. The tags on these fusion proteins can then be used for easy detection and...
8.3K
SNAREs and Membrane Fusion
12.4K
Once a transport vesicle has recognized its target organelle, the vesicular membrane needs to fuse with the target membrane to unload the cargo. Transmembrane proteins called SNAREs present on organelle membranes and their vesicles, mediate vesicle fusion.
SNAREs exist in pairs that symmetrically interact and catalyze the fusion of the lipid bilayers in vesicle and target organelle. v-SNARE in the vesicle membrane are single polypeptide chains that bind to a complementary t-SNARE, composed of 2...
SNAREs exist in pairs that symmetrically interact and catalyze the fusion of the lipid bilayers in vesicle and target organelle. v-SNARE in the vesicle membrane are single polypeptide chains that bind to a complementary t-SNARE, composed of 2...
12.4K
Fusion of Secretory Vesicles with the Plasma Membrane
16.6K
Proteins and neurotransmitters in secretory vesicles can be released from a cell upon vesicle docking, priming, and fusion with the plasma membrane. Vesicles are docked and primed in preparation for the quick exocytosis of their contents in response to a stimulus. The fusion process is mainly carried out by a SNAP Receptor or SNARE complex, consisting of synaptobrevin, syntaxin-1, and SNAP-25.
In 1993, Jim Rothman proposed that the antiparallel pairing of vesicular and transmembrane SNAREs, or...
In 1993, Jim Rothman proposed that the antiparallel pairing of vesicular and transmembrane SNAREs, or...
16.6K
The Unfolded Protein Response
6.3K
The ER is the hub of protein synthesis in a cell. It has robust systems to quality control protein folding and also for degradation of terminally misfolded proteins. Under normal conditions, a small proportion of misfolded proteins that cannot be salvaged need to be transported to the cytoplasm by the ER-associated degradation or ERAD pathways. However, if the ERAD cannot handle the misfolded proteins, the cell activates the unfolded protein response or UPR to adjust the protein folding...
6.3K
Exceptions to the Octet Rule
37.2K
Many covalent molecules have central atoms that do not have eight electrons in their Lewis structures. These molecules fall into three categories:
37.2K


